abstract
stringlengths 383
2.62k
| TLDR
stringlengths 4
250
|
|---|---|
Regularising the parameter matrices of neural networks is ubiquitous in training deep models. Typical regularisation approaches suggest initialising weights using small random values, and to penalise weights to promote sparsity. However, these widely used techniques may be less effective in certain scenarios. Here, we study the Koopman autoencoder model which includes an encoder, a Koopman operator layer, and a decoder. These models have been designed and dedicated to tackle physics-related problems with interpretable dynamics and an ability to incorporate physics-related constraints. However, the majority of existing work employs standard regularisation practices. In our work, we take a step toward augmenting Koopman autoencoders with initialisation and penalty schemes tailored for physics-related settings. Specifically, we propose the "eigeninit" initialisation scheme that samples initial Koopman operators from specific eigenvalue distributions. In addition, we suggest the "eigenloss" penalty scheme that penalises the eigenvalues of the Koopman operator during training. We demonstrate the utility of these schemes on two synthetic data sets: a driven pendulum and flow past a cylinder; and two real-world problems: ocean surface temperatures and cyclone wind fields. We find on these datasets that eigenloss and eigeninit improves the convergence rate by a factor of 2 to 5, and that they reduce the cumulative long-term prediction error by up to a factor of 2.5. Such a finding points to the utility of incorporating similar schemes as an inductive bias in other physics-related deep learning approaches. | Using eigenvalues to regularise and initialise Koopman autoencoders improves performance significantly |
High-resolution images are prevalent in various applications, such as autonomous driving and computer-aided diagnosis. However, training neural networks on such images is computationally challenging and easily leads to out-of-memory errors even on modern GPUs. We propose a simple method, Iterative Patch Selection (IPS), which decouples the memory usage from the input size and thus enables the processing of arbitrarily large images under tight hardware constraints. IPS achieves this by selecting only the most salient patches, which are then aggregated into a global representation for image recognition. For both patch selection and aggregation, a cross-attention based transformer is introduced, which exhibits a close connection to Multiple Instance Learning. Our method demonstrates strong performance and has wide applicability across different domains, training regimes and image sizes while using minimal accelerator memory. For example, we are able to finetune our model on whole-slide images consisting of up to 250k patches (>16 gigapixels) with only 5 GB of GPU VRAM at a batch size of 16. | We propose a simple, memory-efficient method that selects the most salient patches from a high-resolution image and then aggregates them into a global representation for image recognition. |
Biological neural networks are capable of recruiting different sets of neurons to encode different memories. However, when training artificial neural networks on a set of tasks, typically, no mechanism is employed for selectively producing anything analogous to these neuronal ensembles. Further, artificial neural networks suffer from catastrophic forgetting, where the network's performance rapidly deteriorates as tasks are learned sequentially. By contrast, sequential learning is possible for a range of biological organisms. We introduce Learned Context Dependent Gating (LXDG), a method to flexibly allocate and recall `artificial neuronal ensembles', using a particular network structure and a new set of regularization terms. Activities in the hidden layers of the network are modulated by gates, which are dynamically produced during training. The gates are outputs of networks themselves, trained with a sigmoid output activation. The regularization terms we have introduced correspond to properties exhibited by biological neuronal ensembles. The first term penalizes low gate sparsity, ensuring that only a specified fraction of the network is used. The second term ensures that previously learned gates are recalled when the network is presented with input from previously learned tasks. Finally, there is a regularization term responsible for ensuring that new tasks are encoded in gates that are as orthogonal as possible from previously used ones. We demonstrate the ability of this method to alleviate catastrophic forgetting on continual learning benchmarks. When the new regularization terms are included in the model along with Elastic Weight Consolidation (EWC) it achieves better performance on the benchmark `permuted MNIST' than with EWC alone. The benchmark `rotated MNIST' demonstrates how similar tasks recruit similar neurons to the artificial neuronal ensemble. | A method to alleviate catastrophic forgetting in artificial neural networks using learned context dependent activity gates |
Adaptive Moment Estimation (Adam) has been observed to converge faster than stochastic gradient descent (SGD) in practice. However, such an advantage has not been theoretically characterized -- the existing convergence rate of Adam is no better than SGD. We attribute this mismatch between theory and practice to a commonly used assumption: the gradient is globally Lipschitz continuous (called $L$-smooth condition). Specifically, compared to SGD, Adam adaptively chooses a learning rate better suited to the local gradient Lipschitz constant (called local smoothness). This effect becomes
prominent when the local smoothness varies drastically across the domain.
In this paper, we analyze the convergence of Adam under a condition called $(L_0,L_1)$-smooth condition, which allows the gradient Lipschitz constant to change with the gradient norm. This condition has been empirically verified to be more realistic for deep neural networks \citep{zhang2019gradient} than the $L$-smooth condition. Under $(L_0,L_1)$-smooth condition, we establish the convergence for Adam with practical hyperparameters. As such, we argue that Adam can adapt to this local smoothness condition, justifying Adam's \emph{adaptivity}. In contrast, SGD can be arbitrarily slow under this condition. Our result can shed light on the benefit of adaptive gradient methods over non-adaptive ones. | We explain why Adam is faster than SGD through convergence analysis. |
Although machine learning algorithms have achieved state-of-the-art status in image classification, recent studies have substantiated that the ability of the models to learn several tasks in sequence, termed continual learning (CL), often suffers from abrupt degradation of performance from previous tasks. A large body of CL frameworks has been devoted to alleviating this issue. However, we observe that forgetting phenomena in CL are not always unfavorable, especially when there is bias (spurious correlation) in training data. We term such type of forgetting benign forgetting, and categorize detrimental forgetting as malignant forgetting. Based on this finding, our objective in this study is twofold: (a) to discourage malignant forgetting by generating previous representations, and (b) encourage benign forgetting by employing contrastive learning in conjunction with feature-level augmentation. Extensive evaluations of biased experimental setups demonstrate that our proposed method, Learning without Prejudices, is effective for continual unbiased learning. | We propose a novel method, coined Learning without Prejudices, that encourages benign forgetting and regularizes malignant forgetting for continual unbiased learning. |
Shape-based virtual screening is widely used in ligand-based drug design to search chemical libraries for molecules with similar 3D shapes yet novel 2D graph structures compared to known ligands. 3D deep generative models can potentially automate this exploration of shape-conditioned 3D chemical space; however, no existing models can reliably generate geometrically realistic drug-like molecules in conformations with a specific shape. We introduce a new multimodal 3D generative model that enables shape-conditioned 3D molecular design by equivariantly encoding molecular shape and variationally encoding chemical identity. We ensure local geometric and chemical validity of generated molecules by using autoregressive fragment-based generation with heuristic bonding geometries, allowing the model to prioritize the scoring of rotatable bonds to best align the growing conformation to the target shape. We evaluate our 3D generative model in tasks relevant to drug design including shape-conditioned generation of chemically diverse molecular structures and shape-constrained molecular property optimization, demonstrating its utility over virtual screening of enumerated libraries. | We develop a shape-conditioned 3D generative model for ligand-based drug design |
Estimation of shortest-path (SP) distance lies at the heart of network analysis tasks. Along with the rapid emergence of large-scale and complex graphs, approximate SP-representing algorithms that transform a graph into compact and low-dimensional representations are critical for fast and scalable online analysis. Among different approaches, learning-based representation methods have made a breakthrough both in response time and accuracy. Several competitive works in learning-based methods heuristically leverage truncated random walk and optimization on the arbitrary linkage for SP representation learning. However, they have limitations on both exploration range and distance preservation. We propose in this paper an efficient and interpretable SP representation method called Betweenness Centrality-based Distance Resampling (BCDR). First, we prove that betweenness centrality-based random walk can occupy a wider exploration range of distance due to its awareness of high-order path structures. Second, we leverage distance resampling to simulate random shortest paths from original paths and prove that the optimization on such shortest paths preserves distance relations via implicitly decomposing SP distance-based similarity matrix. BCDR yields an average improvement of 25% accuracy and 25-30% query speed, compared to all existing approximate methods when evaluated on a broad class of real-world and synthetic graphs with diverse sizes and structures. | We propose an efficient and interpretable shortest-path representation method for fast, accurate and scalable shortest-path distance queries. |
In multivariate time series forecasting, the most popular strategy for modeling the relationship between multiple time series is the construction of graph, where each time series is represented as a node and related nodes are connected by edges, i.e. spatial-temporal graph neural networks. The graph structure is either given apriori or learned based the similarity between nodes. However, the relationship between multiple time series is typically complicated, for instance, the sum of outflows from upstream nodes may be equal to the inflows of downstream nodes. Such relations widely exist in many real-world multivariate time series forecasting scenarios, yet are far from well studied. In these cases, graph might only be a crude description on the dependency between nodes. To this end, we explore a new framework to model the inter-node relationship in a more precise way based our proposed inductive bias for graphs, Functional Relation Field, where a group of functions parameterized by neural networks are learned to characterize the dependency between multiple time series. These learned functions are versatile: they can then be used to discover the underlying graph structure by identifying the most relevant neighbors of the target node; and on the other hand, the learned functions will form a “field” where the nodes in the backbone prediction networks are enforced to satisfy the constraints defined by these functions. The experiment is conducted on one toy dataset to show our approach can well recover the true constraint relationship between nodes. And two real-world MiniApp calling traffic and road network datasets are also considered with various different backbone networks. Results show that the prediction error can be reduced remarkably with the aid of the proposed functional relation field framework. | Functional Relation Field: A Model-Agnostic Framework for Multivariate Time Series Forecasting |
Advancements in deep learning are often associated with increasing model sizes.
Training and deploying large models require sophisticated hardware and incur
significantly higher costs. Thus, model compression is a widely explored approach
to solving the problem. However, SOTA techniques fall short in one or more
desirable aspects of compression - for instance, pruning does not reduce memory
for training, quantization can only provide up to $32\times$ compression, HashedNet
is cache-inefficient, etc. This paper proposes a model-agnostic, cache-friendly,
and hardware-aware model compression approach: Random Operation Access
Specific Tile (ROAST) hashing. ROAST collapses the parameters by clubbing them
through a lightweight mapping. While clubbing these parameters, ROAST utilizes
cache hierarchies by aligning the memory access pattern with the parameter access
pattern. ROAST is up to $\sim 25 \times$ faster to train and $\sim 50 \times$ faster to infer than the
popular parameter sharing method HashedNet. Additionally, ROAST introduces
global weight sharing, which is empirically and theoretically superior to local
weight sharing in HashedNet, and can be of independent interest. With ROAST, we
can efficiently train and deploy the model using a much smaller memory footprint
($\sim 10 \times - 100 \times$ lesser) in text and image classification tasks | efficient model compression using parameter sharing tuned to underlying hardware and algorithm implementations. |
Prevention of complete and dimensional collapse of representations has recently become a design principle for self-supervised learning (SSL). However, questions remain in our theoretical understanding: When do those collapses occur? What are the mechanisms and causes? We answer these questions by deriving and thoroughly analyzing an analytically tractable theory of SSL loss landscapes. In this theory, we identify the causes of the dimensional collapse and study the effect of normalization and bias. Finally, we leverage the interpretability afforded by the analytical theory to understand how dimensional collapse can be beneficial and what affects the robustness of SSL against data imbalance. | We analytically solve the loss landscape of self-supervised learning and identify the causes of complete and dimensional collapse |
Recent work in machine learning and cognitive science has suggested that understanding causal information is essential to the development of intelligence. One of the key challenges for current machine learning algorithms is modeling and understanding causal overhypotheses: transferable abstract hypotheses about sets of causal relationships. In contrast, even young children spontaneously learn causal overhypotheses, and use these to guide their exploration or to generalize to new situations. This has been demonstrated in a variety of cognitive science experiments using the “blicket detector” environment. We present a causal learning benchmark adapting the “blicket" environment for machine learning agents and evaluate a range of state-of-the-art methods in this environment. We find that although most agents have no problem learning causal structures seen during training, they are unable to learn causal overhypotheses from these experiences, and thus cannot generalize to new settings. | We present a new flexible environment which allows for the evaluation of existing techniques under variable causal overhypotheses |
We present a method to map 2D image observations of a scene to a persistent 3D scene representation, enabling novel view synthesis and disentangled representation of the movable and immovable components of the scene. Motivated by the bird’s-eye-view (BEV) representation commonly used in vision and robotics, we propose conditional neural groundplans, ground-aligned 2D feature grids, as persistent and memory-efficient scene representations. Our method is trained self-supervised from unlabeled multi-view observations using differentiable rendering, and learns to complete geometry and appearance of occluded regions. In addition, we show that we can leverage multi-view videos at training time to learn to separately reconstruct static and movable components of the scene from a single image at test time. The ability to separately reconstruct movable objects enables a variety of downstream tasks using simple heuristics, such as extraction of object-centric 3D representations, novel view synthesis, instance-level segmentation, 3D bounding box prediction, and scene editing. This highlights the value of neural groundplans as a backbone for efficient 3D scene understanding models. | We train a self-supervised model that learns to map a single image to a 3D representation of the scene, with separate components for the immovable and movable 3D regions. |
Prompt tuning, in which a base pretrained model is adapted to each task via conditioning on learned prompt vectors, has emerged as a promising approach for efficiently adapting large language models to multiple downstream tasks. However, existing methods typically learn soft prompt vectors from scratch, and it has not been clear how to exploit the rich cross-task knowledge with prompt vectors in a multitask learning setting. We propose multitask prompt tuning (MPT), which first learns a single transferable prompt by distilling knowledge from multiple task-specific source prompts. We then learn multiplicative low rank updates to this shared prompt to efficiently adapt it to each downstream target task. Extensive experiments on 23 NLP datasets demonstrate that our proposed approach outperforms the state-of-the-art methods, including the full finetuning baseline in some cases, despite only tuning $0.035\%$ as many task-specific parameters. | We propose multitask prompt tuning which learns a single transferable prompt by decomposing and distilling knowledge from multiple task-specific source prompts. |
Unsupervised, category-agnostic, object-centric 3D representation learning for complex scenes remains an open problem in computer vision. While a few recent methods can now discover 3D object radiance fields from a single image without supervision, they are limited to simplistic scenes with objects of a single category, often with a uniform color. This is because they discover objects purely based on appearance cues—objects are made of pixels that look alike. In this work, we propose Movable Object Radiance Fields (MORF), aiming at scaling to complex scenes with diverse categories of objects. Inspired by cognitive science of object learning in babies, MORF learns 3D object representations via movable object inference. During training, MORF first obtains 2D masks of movable objects via a self-supervised movable object segmentation method; it then bridges the gap to 3D object representations via conditional neural rendering in multiple views. During testing, MORF can discover, reconstruct, and move unseen objects from novel categories, all from a single image. Experiments show that MORF extracts accurate object geometry and supports realistic object and scene reconstruction and editing, significantly outperforming the state-of-the-art. | Unsupervised, category-agnostic, object-centric 3D representation learning for complex scenes |
In this paper, we introduce Jointist, an instrument-aware multi-instrument framework that is capable of transcribing, recognizing, and separating multiple musical instruments from an audio clip.
Jointist consists of an instrument recognition module that conditions the other two modules: a transcription module that outputs instrument-specific piano rolls, and a source separation module that utilizes instrument information and transcription results. The joint training of the transcription and source separation modules serves to improve the performance of both tasks. The instrument module is optional and can be directly controlled by human users. This makes Jointist a flexible user-controllable framework.
Our challenging problem formulation makes the model highly useful in the real world given that modern popular music typically consists of multiple instruments. Its novelty, however, necessitates a new perspective on how to evaluate such a model. In our experiments, we assess the proposed model from various aspects, providing a new evaluation perspective for multi-instrument transcription. Subjective listening test shows that Jointist achieves state-of-the-art performance on popular music, outperforming existing multi-instrument transcription models such as MT3. %We also argue that transcription models can be used as a preprocessing module for other music analysis tasks. We conducted experiments on several downstream tasks, and found that the proposed method improved transcription by more than 1 percentage points (ppt.); source separation by 5 SDR, downbeat detection by 1.8 ppt., chord recognition by 1.4 ppt., and key estimation by 1.4 ppt., when utilizing transcription results obtained from Jointist. | Joint training of music transcription and source separation improves the performance of both |
Post hoc auditing of model fairness suffers from potential drawbacks: (1) auditing may be highly sensitive to the test samples chosen; (2) the model and/or its training data may need to be shared with an auditor thereby breaking confidentiality. We address these issues by instead providing a certificate that demonstrates that the learning algorithm itself is fair, and hence, as a consequence, so too is the trained model. We introduce a method to provide a confidential proof of fairness for training, in the context of widely used decision trees, which we term Confidential-PROFITT. We propose novel fair decision tree learning algorithms along with customized zero-knowledge proof protocols to obtain a proof of fairness that can be audited by a third party. Using zero-knowledge proofs enables us to guarantee confidentiality of both the model and its training data. We show empirically that bounding the information gain of each node with respect to the sensitive attributes reduces the unfairness of the final tree. In extensive experiments on the COMPAS, Communities and Crime, Default Credit, and Adult datasets, we demonstrate that a company can use Confidential-PROFITT to certify the fairness of their decision tree to an auditor in less than 2 minutes, thus indicating the applicability of our approach. This is true for both the demographic parity and equalized odds definitions of fairness. Finally, we extend Confidential-PROFITT to apply to ensembles of trees. | We introduce a method to provide a confidential proof of fair training. |
Sharpness-aware minimization (SAM), which searches for flat minima by min-max optimization, has been shown to be useful in improving model generalization. However, since each SAM update requires computing two gradients, its computational cost and training time are both doubled compared to standard empirical risk minimization (ERM). Recent state-of-the-arts reduce the fraction of SAM updates and thus accelerate SAM by switching between SAM and ERM updates randomly or periodically. In this paper, we design an adaptive policy to employ SAM based on the loss landscape geometry. Two efficient algorithms, AE-SAM and AE-LookSAM, are proposed. We theoretically show that AE-SAM has the same convergence rate as SAM. Experimental results on various datasets and architectures demonstrate the efficiency and effectiveness of the adaptive policy. | We design an adaptive policy to employ SAM and propose two efficient algorithms to reduce the fraction of SAM updates. |
By properly utilizing the learned environment model, model-based reinforcement learning methods can improve the sample efficiency for decision-making problems. Beyond using the learned environment model to train a policy, the success of MCTS-based methods shows that directly incorporating the learned environment model as a planner to make decisions might be more effective. However, when action space is of high dimension and continuous, directly planning according to the learned model is costly and non-trivial. Because of two challenges: (1) the infinite number of candidate actions and (2) the temporal dependency between actions in different timesteps. To address these challenges, inspired by Differential Dynamic Programming (DDP) in optimal control theory, we design a novel Policy Optimization with Model Planning (POMP) algorithm, which incorporates a carefully designed Deep Differential Dynamic Programming (D3P) planner into the model-based RL framework. In D3P planner, (1) to effectively plan in the continuous action space, we construct a locally quadratic programming problem that uses a gradient-based optimization process to replace search. (2) To take the temporal dependency of actions at different timesteps into account, we leverage the updated and latest actions of previous timesteps (i.e., step $1, \cdots, h-1$) to update the action of the current step (i.e., step $h$), instead of updating all actions simultaneously. We theoretically prove the convergence rate for our D3P planner and analyze the effect of the feedback term. In practice, to effectively apply the neural network based D3P planner in reinforcement learning, we leverage the policy network to initialize the action sequence and keep the action update conservative in the planning process. Experiments demonstrate that POMP consistently improves sample efficiency on widely used continuous control tasks. Our code is released at https://github.com/POMP-D3P/POMP-D3P. | Directly using the environment model to do the planning might be an efficient way when making decision. We propose a novel POMP algorithm with a D3P planner module to achieve the efficient planning in the continuous action space control problem. |
Early sensory systems in the brain rapidly adapt to fluctuating input statistics, which requires recurrent communication between neurons. Mechanistically, such recurrent communication is often indirect and mediated by local interneurons. In this work, we explore the computational benefits of mediating recurrent communication via interneurons compared with direct recurrent connections. To this end, we consider two mathematically tractable recurrent neural networks that statistically whiten their inputs --- one with direct recurrent connections and the other with interneurons that mediate recurrent communication. By analyzing the corresponding continuous synaptic dynamics and numerically simulating the networks, we show that the network with interneurons is more robust to initialization than the network with direct recurrent connections in the sense that the convergence time for the synaptic dynamics in the network with interneurons (resp. direct recurrent connections) scales logarithmically (resp. linearly) with the spectrum of their initialization. Our results suggest that interneurons are computationally useful for rapid adaptation to changing input statistics. Interestingly, the network with interneurons is an overparameterized solution of the whitening objective for the network with direct recurrent connections, so our results can be viewed as a recurrent neural network analogue of the implicit acceleration phenomenon observed in overparameterized feedforward linear networks. | We show that adding interneurons to a recurrent neural network for statistical whitening accelerates the learning dynamics |
Federated learning (FL) provides general principles for decentralized clients to train a server model collectively without sharing local data. FL is a promising framework with practical applications, but its standard training paradigm requires the clients to backpropagate through the model to compute gradients. Since these clients are typically edge devices and not fully trusted, executing backpropagation on them incurs computational and storage overhead as well as white-box vulnerability. In light of this, we develop backpropagation-free federated learning, dubbed BAFFLE, in which backpropagation is replaced by multiple forward processes to estimate gradients. BAFFLE is 1) memory-efficient and easily fits uploading bandwidth; 2) compatible with inference-only hardware optimization and model quantization or pruning; and 3) well-suited to trusted execution environments, because the clients in BAFFLE only execute forward propagation and return a set of scalars to the server. In experiments, we demonstrate that BAFFLE-trained models can achieve empirically comparable performance to conventional FL models. | BAFFLE is a backpropagation-free and memory-efficient federated learning framework that only executes forward propagation during training. |
One major challenge of disentanglement learning with variational autoencoders is the trade-off between disentanglement and reconstruction fidelity. Previous methods, spreading the conflict of disentanglement and reconstruction in time, will lose the constraint of disentanglement when expanding the information bottleneck, which causes the information diffusion problem. To tackle this issue, we present a novel decremental variational autoencoder with disentanglement-invariant transformations to spread the conflict on multiple latent spaces, termed DeVAE, for balancing disentanglement and reconstruction fidelity by decreasing the information bottleneck of diverse latent spaces gradually. Benefiting from the multiple latent spaces, DeVAE allows simultaneous optimization of multiple objectives to optimize reconstruction while keeping the constraint of disentanglement, avoiding information diffusion. DeVAE is also compatible with large models with high-dimension latent space. Experimental results on dSprites and Shapes3D that DeVAE achieves the best performance on both disentanglement and reconstruction. | We present a novel decremental variational autoencoder with disentanglement-invariant transformations, termed DeVAE, for balancing disentanglement and reconstruction fidelity. |
Artificial and biological agents are unable to learn given completely random and unstructured data. The structure of data is encoded in the distance or similarity relationships between data points. In the context of neural networks, the neuronal activity within a layer forms a representation reflecting the transformation that the layer implements on its inputs. In order to utilize the structure in the data in a truthful manner, such representations should reflect the input distances and thus be continuous and isometric. Supporting this statement, recent findings in neuroscience propose that generalization and robustness are tied to neural representations being continuously differentiable. However, in machine learning, most algorithms lack robustness and are generally thought to rely on aspects of the data that differ from those that humans use, as is commonly seen in adversarial attacks.
During cross-entropy classification, the metric and structural properties of network representations are usually broken both between and within classes. This side effect from training can lead to instabilities under perturbations near locations where such structure is not preserved. One of the standard solutions to obtain robustness is to train specifically by introducing perturbations in the training data. This leads to networks that are particularly robust to specific training perturbations but not necessarily to general perturbations. While adding ad hoc regularization terms to improve robustness has become common practice, to our knowledge, forcing representations to preserve the metric structure of the input data as a stabilising mechanism has not yet been introduced.
In this work, we train neural networks to perform classification while simultaneously maintaining the metric structure within each class, leading to continuous and isometric within-class representations. We show that such network representations turn out to be a beneficial component for making accurate and robust inferences about the world. By stacking layers with this property we provide the community with an network architecture that facilitates hierarchical manipulation of internal neural representations. Finally, we verify that our isometric regularization term improves the robustness to adversarial attacks on MNIST. | Adding Isometric regularization to classification loss, enforces continuous representations which improve robustness against adversarial attacks. |
The remarkable performance gains realized by large pretrained models, e.g., GPT-3, hinge on the massive amounts of data they are exposed to during training. Analogously, distilling such large models to compact models for efficient deployment also necessitates a large amount of (labeled or unlabeled) training data. In this paper, we propose the teacher-guided training (TGT) framework for training a high-quality compact model that leverages the knowledge acquired by pretrained generative models, while obviating the need to go through a large volume of data. TGT exploits the fact that the teacher has acquired a good representation of the underlying data domain, which typically corresponds to a much lower dimensional manifold than the input space. Furthermore, we can use the teacher to explore input space more efficiently through sampling or gradient-based methods; thus, making TGT especially attractive for limited data or long-tail settings. We formally capture this benefit of proposed data-domain exploration in our generalization bounds. We find that TGT can improve accuracy on several image classification benchmarks as well as a range of text classification and retrieval tasks. | We propose and theoretically analyze a novel way to improve the training efficiency of compact student models that better leverages the knowledge of pretrained generative (teacher) models compared to standard distillation methods. |
Chain-of-thought prompting has demonstrated remarkable performance on various natural language reasoning tasks. However, it tends to perform poorly on tasks which requires solving problems harder than the exemplars shown in the prompts. To overcome this challenge of easy-to-hard generalization, we propose a novel prompting strategy, least-to-most prompting. The key idea in this strategy is to break down a complex problem into a series of simpler subproblems and then solve them in sequence. Solving each subproblem is facilitated by the answers to previously solved subproblems. Our experimental results on tasks related to symbolic manipulation, compositional generalization, and math reasoning reveal that least-to-most prompting is capable of generalizing to more difficult problems than those seen in the prompts. A notable finding is that when the GPT-3 code-davinci-002 model is used with least-to-most prompting, it can solve the compositional generalization benchmark SCAN in any split (including length split) with an accuracy of at least 99\% using just 14 exemplars, compared to only 16\% accuracy with chain-of-thought prompting. This is particularly noteworthy because neural-symbolic models in the literature that specialize in solving SCAN are trained on the entire training set containing over 15,000 examples. We have included prompts for all the tasks in the Appendix. | We propose a novel prompting strategy, least-to-most prompting, that enables large language models to achieve easy-to-hard generalization |
Graphs are ubiquitous and are often used to understand the dynamics of a system. Probabilistic Graphical Models comprising Bayesian and Markov networks, and Conditional Independence graphs are some of the popular graph representation techniques. They can model relationships between features (nodes) together with the underlying distribution. Although theoretically these models can represent very complex dependency functions, in practice often simplifying assumptions are made due to computational limitations associated with graph operations. This work introduces Neural Graphical Models (NGMs) which attempt to represent complex feature dependencies with reasonable computational costs. Specifically, given a graph of feature relationships and corresponding samples, we capture the dependency structure between the features along with their complex function representations by using neural networks as a multi-task learning framework. We provide efficient learning, inference and sampling algorithms for NGMs. Moreover, NGMs can fit generic graph structures including directed, undirected and mixed-edge graphs as well as support mixed input data types. We present empirical studies that show NGMs' capability to represent Gaussian graphical models, inference analysis of a lung cancer data and extract insights from a real world infant mortality data provided by CDC. | A neural network based graphical model with efficient learning, inference and sampling algorithms |
The recent push for explainable artificial intelligence (XAI) has given rise to extensive work toward understanding the inner workings of neural networks. Much of that work, however, has focused on manipulating input data feeding the network to assess their effect on network output. It is shown in this study that XAI can benefit from investigating the network node, the most fundamental unit of neural networks. Whereas studies on XAI have mostly benefited from a focus on manipulating input data, assessing patterns in node weights may prove equally beneficial, if not more significant, especially when realizing that weight values may not be as random as previously thought. A manipulated, a contrived, and a real dataset were used in this study. Datasets were run on convolutional and deep neural network models. Node rank stability was the central construct to investigate neuronal patterns in this study. Rank stability was defined as the number of epochs wherein nodes held their rank in terms of weight value compared to their rank at the last epoch, when the model reached convergence, or stability (defined in this study as accuracy $\geq$ 0.90). Findings indicated that neural networks behaved like a decision tree, in that rank stability increased as weight absolute values increased. Decision tree behavior may assist in more efficient pruning algorithms, which may produce distilled models simpler to explain to technical and non-technical audiences. | This work assessed node weight patterns toward explaining artificial intelligence systems. |
Decentralized stochastic gradient descent (D-SGD) allows collaborative learning on massive devices without controlling of a central server. Existing theory suggests that the decentralization degrades the generalizability, which conflicts with experimental results in the large-batch settings that D-SGD generalize better than centralized SGD (C-SGD). This work presents new theory that reconciles the conflict between the two perspectives. We prove that D-SGD introduces an implicit regularization that simultaneously penalizes (1) the sharpness of the learned minima and (2) the consensus distance between the consensus model and local models. We then prove that the implicit regularization is amplified in the large-batch settings when the linear scaling rule is applied. We further analyze the escaping efficiency of D-SGD, which suggests that D-SGD favors super-quadratic flat minima. Experiments are in full agreement with our theory. The code will be released publicly. To our best knowledge, this is the first work on the implicit regularization and escaping efficiency of D-SGD. | D-SGD introduces an implicit regularization that penalizes the learned minima’s sharpness. |
Large language models (LLMs) are useful in many NLP tasks and become more capable with size, scaling to over 100 billion parameters. With the release of BLOOM-176B and OPT-175B, everyone can download pretrained models of this scale. Still, using a pre-trained 100B+ model requires high-end hardware, making it inaccessible to most researchers. Recent studies in memory-efficient training (e.g. offloading) could alleviate these costs, but they do not cover important use cases of LLMs, such as autoregressive inference. In this work, we investigate methods for cost-efficient inference of large language models, comparing local and distributed strategies. We observe that a large enough model (100B+) could run efficiently on geodistributed devices in a consumer-grade network, for example by connecting existing compute resources of multiple research groups or pooling under-utilized compute from multiple cloud regions. To run LLMs in this unconventional setting, we develop a fault-tolerant algorithm for inferencing language models. We propose Petals - a decentralized system for running LLMs - and show that it can run BLOOM-176B over the Internet over $10\times$ faster than offloading for sequential generation. We evaluate the performance of our system in both simulated conditions and an actual distributed system spanning two continents. The design of Petals allows participants to inference, and fine-tune, or inference fine-tuned models simultaneously without affecting each other's results. | We propose a practical algorithm for running large language models by pooling together weak geographically distributed devices. Our system can inference BLOOM-176B over the Internet more than 10x faster compared to RAM offloading. |
Selecting informative data points for expert feedback can significantly improve the performance of anomaly detection in various contexts, such as medical diagnostics or fraud detection. In this paper, we determine a set of conditions under which the ranking of anomaly scores generalizes from labeled queries to unlabeled data. Inspired by these conditions, we propose a new querying strategy for active anomaly detection that leads to systematic improvements over current approaches for this problem. It selects a diverse set of data points for labeling, achieving high data coverage with a limited budget. These labeled data points provide weak supervision to the unsupervised anomaly detection problem. However, correctly identifying anomalies requires an estimate of the fraction of anomalies in the data. We show how this anomaly rate can be estimated from the query set by importance-weighting, removing the associated bias due to the non-uniform sampling procedure. Extensive experiments on image, tabular, and video data sets show that our approach results in state-of-the-art active anomaly detection performance. | A new active learning approach for deep anomaly detection that leads to leads to systematic improvements over current approaches. |
In this work, we show that recently proposed quadratic models capture optimization and generalization properties of wide neural networks that cannot be captured by linear models. In particular, we prove that quadratic models for shallow ReLU networks exhibit the "catapult phase" from Lewkowycz et al. (2020) that arises when training such models with large learning rates. We then empirically show that the behaviour of quadratic models parallels that of neural networks in generalization, especially in the catapult phase regime. Our analysis further demonstrates that quadratic models are an effective tool for analysis of neural networks. | Quadratic models capture properties of wide neural networks in both optimization and generalization. |
Test-time adaptation (TTA) aims to adapt a trained classifier using online unlabeled test data only, without any information related to the training procedure. Most existing TTA methods adapt the trained classifier using the classifier's prediction on the test data as pseudo-label.
However, under test-time domain shift, accuracy of the pseudo labels cannot be guaranteed, and thus the TTA methods often encounter performance degradation at the adapted classifier. To overcome this limitation, we propose a novel test-time adaptation method, called Test-time Adaptation via Self-Training with nearest neighbor information (TAST), which is composed of the following procedures: (1) adds trainable adaptation modules on top of the trained feature extractor; (2) newly defines a pseudo-label distribution for the test data by using the nearest neighbor information; (3) trains these modules only a few times during test time to match the nearest neighbor-based pseudo label distribution and a prototype-based class distribution for the test data; and (4) predicts the label of test data using the average predicted class distribution from these modules. The pseudo-label generation is based on the basic intuition that a test data and its nearest neighbor in the embedding space are likely to share the same label under the domain shift. By utilizing multiple randomly initialized adaptation modules, TAST extracts useful information for the classification of the test data under the domain shift, using the nearest neighbor information. TAST showed better performance than the state-of-the-art TTA methods on two standard benchmark tasks, domain generalization, namely VLCS, PACS, OfficeHome, and TerraIncognita, and image corruption, particularly CIFAR-10/100C.
| This work presents a simple and efficient test-time adaptation method to adapt trained classifiers by utilizing an ensemble of adaptation modules and self-training with nearest neighbor information. |
Mapping biological mechanisms in cellular systems is a fundamental step in early-stage drug discovery that serves to generate hypotheses on what disease-relevant molecular targets may effectively be modulated by pharmacological interventions. With the advent of high-throughput methods for measuring single-cell gene expression under genetic perturbations, we now have effective means for generating evidence for causal gene-gene interactions at scale. However, inferring graphical networks of the size typically encountered in real-world gene-gene interaction networks is difficult in terms of both achieving and evaluating faithfulness to the true underlying causal graph. Moreover, standardised benchmarks for comparing methods for causal discovery in perturbational single-cell data do not yet exist. Here, we introduce CausalBench - a comprehensive benchmark suite for evaluating network inference methods on large-scale perturbational single-cell gene expression data. CausalBench introduces several biologically meaningful performance metrics and operates on two large, curated and openly available benchmark data sets for evaluating methods on the inference of gene regulatory networks from single-cell data generated under perturbations. With real-world datasets consisting of over 200000 training samples under interventions, CausalBench could potentially help facilitate advances in causal network inference by providing what is - to the best of our knowledge - the largest openly available test bed for causal discovery from real-world perturbation data to date. | We introduce CausalBench - a comprehensive benchmark suite for evaluating network inference methods on perturbational single-cell gene expression data. |
We study regret minimization for reinforcement learning (RL) in Latent Markov Decision Processes (LMDPs) with context in hindsight. We design a novel model-based algorithmic framework which can be instantiated with both a model-optimistic and a value-optimistic solver. We prove an $\widetilde{O}\left(\sqrt{M \Gamma S A K}\right)$ regret bound where $M$ is the number of contexts, $S$ is the number of states, $A$ is the number of actions, $K$ is the number of episodes, and $\Gamma \le S$ is the maximum transition degree of any state-action pair. The regret bound only scales logarithmically with the planning horizon, thus yielding the first (nearly) horizon-free regret bound for LMDP. Key in our proof is an analysis of the total variance of alpha vectors, which is carefully bounded by a recursion-based technique. We complement our positive result with a novel $\Omega\left(\sqrt{M S A K}\right)$ regret lower bound with $\Gamma = 2$, which shows our upper bound minimax optimal when $\Gamma$ is a constant. Our lower bound relies on new constructions of hard instances and an argument based on the symmetrization technique from theoretical computer science, both of which are technically different from existing lower bound proof for MDPs, and thus can be of independent interest. | We studied RL for Latent MDPs under the episodic, context in hindsight setting. A SOTA upperbound and the first lowerbound were presented. |
We propose a novel knowledge distillation (KD) method to selectively instill teacher knowledge into a student model motivated by situations where the student's capacity is significantly smaller than that of the teachers. In vanilla KD, the teacher primarily sets a predictive target for the student to follow, and we posit that this target is overly optimistic due to the student's lack of capacity. We develop a novel scaffolding scheme where the teacher, in addition to setting a predictive target, also scaffolds the student's prediction by censoring hard-to-learn examples. Scaffolding utilizes the same information as the teacher's soft-max predictions as inputs, and in this sense, our proposal can be viewed as a natural variant of vanilla KD. We show on synthetic examples that censoring hard-examples leads to smoothening the student's loss landscape so that the student encounters fewer local minima. As a result, it has good generalization properties. Against vanilla KD, we achieve improved performance and are comparable to more intrusive techniques that leverage feature matching on benchmark datasets.
| We develop a novel KD scheme where the teacher scaffolds the student's prediction on hard-to-learn examples. It smoothens student's loss landscape so that the student encounters fewer local minima. As a result it has good generalization properties. |
Clustering is a widely deployed unsupervised learning tool. Model-based clustering is a flexible framework to tackle data heterogeneity when the clusters have different shapes. Likelihood-based inference for mixture distributions often involves non-convex and high-dimensional objective functions, imposing difficult computational and statistical challenges. The classic expectation-maximization (EM) algorithm is a computationally thrifty iterative method that maximizes a surrogate function minorizing the log-likelihood of observed data in each iteration, which however suffers from bad local maxima even in the special case of the standard Gaussian mixture model with common isotropic covariance matrices. On the other hand, recent studies reveal that the unique global solution of a semidefinite programming (SDP) relaxed $K$-means achieves the information-theoretically sharp threshold for perfectly recovering the cluster labels under the standard Gaussian mixture model. In this paper, we extend the SDP approach to a general setting by integrating cluster labels as model parameters and propose an iterative likelihood adjusted SDP (iLA-SDP) method that directly maximizes the \emph{exact} observed likelihood in the presence of data heterogeneity. By lifting the cluster assignment to group-specific membership matrices, iLA-SDP avoids centroids estimation -- a key feature that allows exact recovery under well-separateness of centroids without being trapped by their adversarial configurations. Thus iLA-SDP is less sensitive than EM to initialization and more stable on high-dimensional data. Our numeric experiments demonstrate that iLA-SDP can achieve lower mis-clustering errors over several widely used clustering methods including $K$-means, SDP and EM algorithms. | We propose an iterative likelihood adjusted SDP for clustering under data heterogeneity. |
Generative models with discrete latent representations have recently demonstrated an impressive ability to learn complex high-dimensional data distributions. However, their performance relies on a long sequence of tokens per instance and a large number of codebook entries, resulting in long sampling times and considerable computation to fit the categorical posterior.
To address these issues, we propose the Masked Vector Quantization (MVQ) framework which increases the representational capacity of each code vector by learning mask configurations via a stochastic winner-takes-all training regime called Multiple Hypotheses Dropout (MH-Dropout). On ImageNet 64$\times$64, reduces FID in existing vector quantization architectures by up to $68\%$ at 2 tokens per instance and $57\%$ at 5 tokens. These improvements widen as codebook entries is reduced and allows for $7\textup{-}45\times$ speed-up in token sampling during inference. As an additional benefit, we find that smaller latent spaces lead to MVQ identifying transferable visual representations where multiple can be smoothly combined. | We proposed Masked Vector Quantization, a novel variant of Vector Quantization, which increases the representational capacity of each code vector by learning mask configurations via winner-takes-all training regime called Multiple Hypotheses Dropout. |
Building scalable models to learn from diverse, multimodal data remains an open challenge.
For vision-language data, the dominant approaches are based on contrastive learning objectives that train a separate encoder for each modality. While effective, contrastive learning approaches introduce sampling bias depending on the data augmentations used, which can degrade performance on downstream tasks. Moreover, these methods are limited to paired image-text data, and cannot leverage widely-available unpaired data. In this paper, we investigate whether a large multimodal model trained purely via masked token prediction, without using modality-specific encoders or contrastive learning, can learn transferable representations for downstream tasks. We propose a simple and scalable network architecture, the Multimodal Masked Autoencoder (M3AE), which learns a unified encoder for both vision and language data via masked token prediction. We provide an empirical study of M3AE trained on a large-scale image-text dataset, and find that M3AE is able to learn generalizable representations that transfer well to downstream tasks. Surprisingly, we find that M3AE benefits from a higher text mask ratio (50-90%), in contrast to BERT whose standard masking ratio is 15%, due to the joint training of two data modalities. We also provide qualitative analysis showing that the learned representation incorporates meaningful information from both image and language. Lastly, we demonstrate the scalability of M3AE with larger model size and training time, and its flexibility to train on both paired image-text data as well as unpaired data.
| Multimodal Transformers for vision-language representation learning trained with masked token prediction. |
Explanation is a key component for the adoption of reinforcement learning (RL) in many real-world decision-making problems. In the literature, the explanation is often provided by saliency attribution to the features of the RL agent's state. In this work, we propose a complementary approach to these explanations, particularly for offline RL, where we attribute the policy decisions of a trained RL agent to the trajectories encountered by it during training. To do so, we encode trajectories in offline training data individually as well as collectively (encoding a set of trajectories). We then attribute policy decisions to a set of trajectories in this encoded space by estimating the sensitivity of the decision with respect to that set. Further, we demonstrate the effectiveness of the proposed approach in terms of quality of attributions as well as practical scalability in diverse environments that involve both discrete and continuous state and action spaces such as grid-worlds, video games (Atari) and continuous control (MuJoCo). We also conduct a human study on a simple navigation task to observe how their understanding of the task compares with data attributed for a trained RL policy. | This work focuses on idea of explaining actions of offline RL agent by attributing the actions to trajectories encountered during the training. |
Long-term engagement is preferred over immediate engagement in sequential recommendation as it directly affects product operational metrics such as daily active users (DAUs) and dwell time. Meanwhile, reinforcement learning (RL) is widely regarded as a promising framework for optimizing long-term engagement in sequential recommendation. However, due to expensive online interactions, it is very difficult for RL algorithms to perform state-action value estimation, exploration and feature extraction when optimizing long-term engagement. In this paper, we propose ResAct which seeks a policy that is close to, but better than, the online-serving policy. In this way, we can collect sufficient data near the learned policy so that state-action values can be properly estimated, and there is no need to perform online exploration. ResAct optimizes the policy by first reconstructing the online behaviors and then improving it via a Residual Actor. To extract long-term information, ResAct utilizes two information-theoretical regularizers to confirm the expressiveness and conciseness of features. We conduct experiments on a benchmark dataset and a large-scale industrial dataset which consists of tens of millions of recommendation requests. Experimental results show that our method significantly outperforms the state-of-the-art baselines in various long-term engagement optimization tasks. | We propose a novel paradigm to reinforce long-term engagement in sequential recommendation. |
We study the generalization abilities of language models when translating natural language into formal specifications with complex semantics. In particular, we fine-tune language models on three datasets consisting of English sentences and their corresponding formal representation: 1) regular expressions (regex), frequently used in programming and search; 2) First-order logic (FOL), commonly used in software verification and theorem proving; and 3) linear-time temporal logic (LTL), which forms the basis for industrial hardware specification languages. Our experiments show that, in these diverse domains, the language models maintain their generalization capabilities from pre-trained knowledge of natural language to generalize, e.g., to new variable names or operator descriptions. Additionally, they achieve competitive performance, and even outperform the state-of-the-art for translating into regular expressions, with the benefits of being easy to access, efficient to fine-tune, and without a particular need for domain-specific reasoning. | We study the generalization abilities of language models when translating natural language into formal specifications with complex semantics. |
Contrastive learning, a self-supervised learning method that can learn representations from unlabeled data, has been developed promisingly. Many methods of contrastive learning depend on data augmentation techniques, which generate different views from the original signal. However, tuning policies and hyper-parameters for more effective data augmentation methods in contrastive learning is often time and resource-consuming. Researchers have designed approaches to automatically generate new views for some input signals, especially on the image data. But the view-learning method is not well developed for time-series data. In this work, we propose a simple but effective module for automating view generation for time-series data in contrastive learning, named learning views for time-series data (LEAVES). The proposed module learns the hyper-parameters for augmentations using adversarial training in contrastive learning. We validate the effectiveness of the proposed method using multiple time-series datasets. The experiments demonstrate that the proposed method is more effective in finding reasonable views and performs downstream tasks better than the baselines, including manually tuned augmentation-based contrastive learning methods and SOTA methods. | We propose a simple but effective method to automatrically learn views for time-series data in contrastive learning |
Dense prediction tasks are a fundamental class of problems in computer vision. As supervised methods suffer from high pixel-wise labeling cost, a few-shot learning solution that can learn any dense task from a few labeled images is desired. Yet, current few-shot learning methods target a restricted set of tasks such as semantic segmentation, presumably due to challenges in designing a general and unified model that is able to flexibly and efficiently adapt to arbitrary tasks of unseen semantics. We propose Visual Token Matching (VTM), a universal few-shot learner for arbitrary dense prediction tasks. It employs non-parametric matching on patch-level embedded tokens of images and labels that encapsulates all tasks. Also, VTM flexibly adapts to any task with a tiny amount of task-specific parameters that modulate the matching algorithm. We implement VTM as a powerful hierarchical encoder-decoder architecture involving ViT backbones where token matching is performed at multiple feature hierarchies. We experiment VTM on a challenging variant of Taskonomy dataset and observe that it robustly few-shot learns various unseen dense prediction tasks. Surprisingly, it is competitive with fully supervised baselines using only 10 labeled examples of novel tasks ($0.004\%$ of full supervision) and sometimes outperforms using $0.1\%$ of full supervision. Codes are available at https://github.com/GitGyun/visual_token_matching. | a universal few-shot learner for general dense prediction tasks |
While current node classification methods for graphs have enabled significant progress in many applications, they rely on abundant labeled nodes for training. In many real-world datasets, nodes for some classes are always scarce, thus current algorithms are ill-equipped to handle these few-shot node classes. Some meta learning approaches for graphs have demonstrated advantages in tackling such few-shot problems, but they disregard the impact of node importance on a task. Being exclusive to graph data, the dependencies between nodes convey vital information for determining the importance of nodes in contrast to node features only, which poses unique challenges here. In this paper, we investigate the effect of node importance in node classification meta learning tasks. We first theoretically analyze the influence of distinguishing node importance on the lower bound of the model accuracy. Then, based on the theoretical conclusion, we propose a novel Node Importance Meta Learning architecture (NIML) that learns and applies the importance score of each node for meta learning. Specifically, after constructing an attention vector based on the interaction between a node and its neighbors, we train an importance predictor in a supervised manner to capture the distance between node embedding and the expectation of same-class embedding. Extensive experiments on public datasets demonstrate the state-of-the-art performance of NIML on few-shot node classification problems. | This paper focuses on the few-shot node classification problem in graph; theoretically studies the influence of node importance on the model accuracy and proposes a node importance calculation method to implement on meta learning GNNs. |
This work studies training one-hidden-layer overparameterized ReLU networks via gradient descent in the neural tangent kernel (NTK) regime, where, differently from the previous works, the networks' biases are trainable and are initialized to some constant rather than zero. The tantalizing benefit of such initialization is that the neural network will provably have sparse activation pattern before, during and after training, which can enable fast training procedures and, therefore, reduce the training cost. The first set of results of this work characterize the convergence of the network's gradient descent dynamics. The required width is provided to ensure gradient descent can drive the training error towards zero in linear rate. The contribution over previous work is that not only the bias is allowed to be updated by gradient descent under our setting but also a finer analysis is given such that the required width to ensure the network's closeness to its NTK is improved. Secondly, the networks' generalization bound after training is provided. A width-sparsity dependence is presented which yields sparsity-dependent localized Rademacher complexity and same-as-previous (ignoring logarithmic factors) generalization bound. Up to our knowledge, this is the first sparsity-dependent generalization result via localized Rademacher complexity. As a by-product, if the bias initialization is chosen to be zero, the width requirement improves the previous bound for the shallow networks' generalization. Lastly, since the generalization bound has dependence on the smallest eigenvalue of the limiting NTK and the bounds from previous works yield vacuous generalization, this work further studies the least eigenvalue of the limiting NTK. Surprisingly, while it is not shown that trainable biases are necessary, trainable bias helps to identify a nice data-dependent region where a much finer analysis of the NTK's smallest eigenvalue can be conducted, which leads to a much sharper lower bound than the previously known worst-case bound and, consequently, a non-vacuous generalization bound. Experimental evaluation is provided to evaluate our results. | We study convergence and generalization of training one-hidden-layer neural networks with sparse activation. |
This work studies the threats of adversarial attack on multivariate probabilistic forecasting models and viable defense mechanisms. Our studies discover a new attack pattern that negatively impact the forecasting of a target time series via making strategic, sparse (imperceptible) modifications to the past observations of a small number of other time series. To mitigate the impact of such attack, we have developed two defense strategies. First, we extend a previously developed randomized smoothing technique in classification to multivariate forecasting scenarios. Second, we develop an adversarial training algorithm that learns to create adversarial examples and at the same time optimizes the forecasting model to improve its robustness against such adversarial simulation. Extensive experiments on real-world datasets confirm that our attack schemes are powerful and our defense algorithms are more effective compared with baseline defense mechanisms.
| Designs of Adversarial Attacks and Defense Mechanisms for Multivariate Forecasting Models |
Finding (local) Nash equilibria in two-player differentiable games is a classical problem in game theory with important relevance in machine learning. We propose double Follow-the-Ridge (double-FTR), an algorithm that locally converges to and only to local Nash equilibria in general-sum two-player differentiable games. To our knowledge, double-FTR is the first algorithm with such guarantees for general-sum games. Furthermore, we show that by varying its preconditioner, double-FTR leads to a broader family of algorithms with the same convergence guarantee. In addition, double-FTR avoids oscillation near equilibria due to the real-eigenvalues of its Jacobian at fixed points.
Empirically, we validate the double-FTR algorithm on a range of simple zero-sum and general sum games, as well as simple Generative Adversarial Network (GAN) tasks. | We propose double Follow-the-Ridge (double-FTR), an algorithm with local convergence guarantee to differential Nash equilibria in general-sum two-player differential games. |
World knowledge exists in both structured (tables, knowledge graphs) and unstructured forms (texts). Recently, there have been extensive research efforts in the integration of structured factual knowledge and unstructured textual knowledge. However, most studies focus on incorporating static factual knowledge into pre-trained language models, while there is less work on enhancing temporal knowledge graph embedding using textual knowledge. Existing integration approaches can not apply to temporal knowledge graphs (tKGs) since they often assume knowledge embedding is time-invariant. In fact, the entity embedding in tKG embedding models usually evolves over time, which poses the challenge of aligning temporally relevant textual information with entities. To this end, we propose Enhanced Temporal Knowledge Embeddings with Contextualized Language Representations (ECOLA), which uses tKG quadruple as an implicit measure to temporally align textual data and the time-evolving entity representations and uses a novel knowledge-text prediction task to inject textual information into temporal knowledge embedding. ECOLA jointly optimizes the knowledge-text prediction objective and the temporal knowledge embedding objective, and thus, can simultaneously take full advantage of textual and structured knowledge. Since existing datasets do not provide tKGs with aligned textual data, we introduce three new datasets for training and evaluating ECOLA. Experimental results on the temporal knowledge graph completion task show that ECOLA outperforms state-of-the-art tKG embedding models by a large margin. | We use textual knowledge to enhance the time-evolving knowledge graph embedding, while preserving its temporal nature. |
Groups with complex set intersection relations are a natural way to model a wide array of data, from the formation of social groups to the complex protein interactions which form the basis of biological life. While graphs are a natural way to represent complex networks and are well studied, typical approaches to modeling group membership using graphs are lossy. Hypergraphs are a more natural way to represent such ``higher order'' relationships, but efforts to apply machine learning techniques to hypergraph structured datasets have been limited thus far. In this paper, we address the problem of link prediction in knowledge hypergraphs as well as regular hypergraphs and develop a novel, simple, and effective optimization architecture to solve this task. Additionally, we study how integrating data from node-level labels can improve the results of our system. Our self-supervised approach achieves significant improvement over state of the art results on several hyperedge prediction and knowledge hypergraph completeion benchmarks. | A new state-of-the-art hyperedge prediction framework for knowledge hypergraphs as well as regular hypergraphs |
The Generative Flow Network is a probabilistic framework where an agent learns a stochastic policy for object generation, such that the probability of generating an object is proportional to a given reward function. Its effectiveness has been shown in discovering high-quality and diverse solutions, compared to reward-maximizing reinforcement learning-based methods. Nonetheless, GFlowNets only learn from rewards of the terminal states, which can limit its applicability. Indeed, intermediate rewards play a critical role in learning, for example from intrinsic motivation to provide intermediate feedback even in particularly challenging sparse reward tasks. Inspired by this, we propose Generative Augmented Flow Networks (GAFlowNets), a novel learning framework to incorporate intermediate rewards into GFlowNets. We specify intermediate rewards by intrinsic motivation to tackle the exploration problem in sparse reward environments. GAFlowNets can leverage edge-based and state-based intrinsic rewards in a joint way to improve exploration. Based on extensive experiments on the GridWorld task, we demonstrate the effectiveness and efficiency of GAFlowNet in terms of convergence, performance, and diversity of solutions. We further show that GAFlowNet is scalable to a more complex and large-scale molecule generation domain, where it achieves consistent and significant performance improvement. | We propose a novel GFlowNet learning framework to incorporate intermediate rewards represented by intrinsic motivation to improve exploration. |
Representation learning for proteins is an emerging area in geometric deep learning. Recent works have factored in both the relational (atomic bonds) and the geometric aspects (atomic positions) of the task, notably bringing together graph neural networks (GNNs) with neural networks for point clouds. The equivariances and invariances to geometric transformations (group actions such as rotations and translations) so far treats large molecules as rigid structures. However, in many important settings, proteins can co-exist as an ensemble of multiple stable conformations. The conformations of a protein, however, cannot be described as input-independent transformations of the protein: Two proteins may require different sets of transformations in order to describe their set of viable conformations. To address this limitation, we introduce the concept of conditional transformations (CT). CT can capture protein structure, while respecting the restrictions posed by constraints on dihedral (torsion) angles and steric repulsions between atoms. We then introduce a Markov chain Monte Carlo framework to learn representations that are invariant to these conditional transformations. Our results show that endowing existing baseline models with these conditional transformations helps improve their performance without sacrificing computational cost. | We propose the conditional invariance (CI) framework, which captures input-dependent transformation invariances as an add-on to existing neural network methods. We augment existing protein GNNs with CI to learn conformer invariant representations. |
The standard Gaussian Process (GP) is formulated to only consider a single output sample for each input in the training set. Datasets for subjective tasks, such as spoken language assessment, may be annotated with output labels from multiple human raters for each input. This paper proposes to generalise the GP to allow for multiple output samples per input in the training set. This differs from a multi-output GP, because all output samples are from the same task here. The output density function is formulated to be the joint likelihood of observing all output samples. Through this, the hyper-parameters are optimised using a criterion that is similar to minimising a Kullback-Leibler divergence. The test set predictions are inferred fairly similarly to a standard GP, with a key difference being in the optimised hyper-parameters. This approach is evaluated on spoken language assessment tasks, using the public speechocean762 dataset and an internal Tamil language dataset. The results show that by using the proposed method, the GP is able to compute a test set output distribution that is more similar to the collection of reference outputs annotated by multiple human raters. | This paper proposes to extend the Gaussian process framework to allow for multiple output samples for each input from the same task in the training set. |
Generalized Eigenvalue Problems (GEPs) encompass a range of interesting scientific computing problems. Canonical Correlation Analysis (CCA) and Partial Least Squares (PLS) are two such examples of GEPs which are often used to learn representations of multiview data. Development of efficient stochastic approaches to these problems would allow them to scale to large datasets. Furthermore, existing deep learning based extensions of CCA require large minibatch sizes in the stochastic setting to achieve good performance. Inspired by recent formulations of Principal Components Analysis and GEPs as games with differentiable utilities, we develop an alternative game theoretic approach to solving GEPs in which all constraints are softly enforced by Lagrange multipliers. We show that our approach shares much of the theoretical grounding of the previous game theoretic approaches but has fewer hyperparameters, is faster to converge, and permits extension to general function approximators like neural networks for certain GEPs including CCA. We demonstrate the effectiveness of our method for solving GEPs using canonical multiview datasets and demonstrate state-of-the-art performance for the Deep CCA problem for multiview representation learning. | A new approach to solving Generalized Eigenvalue Problems in the stochastic setting extended to Deep Canonical Correlation Analysis with state-of-the-art results for stochastic minibatches |
A default assumption in reinforcement learning and optimal control is that experience arrives at discrete time points on a fixed clock cycle. Many applications, however, involve continuous systems where the time discretization is not fixed but instead can be managed by a learning algorithm. By analyzing Monte-Carlo value estimation for LQR systems in both finite-horizon and infinite-horizon settings, we uncover a fundamental trade-off between approximation and statistical error in value estimation. Importantly, these two errors behave differently with respect to time discretization, which implies that there is an optimal choice for the temporal resolution that depends on the data budget. These findings show how adapting the temporal resolution can provably improve value estimation quality in LQR systems from finite data. Empirically, we demonstrate the trade-off in numerical simulations of LQR instances and several non-linear environments. | By analyzing Monte-Carlo value estimation for LQR systems we uncover a fundamental trade-off between approximation and statistical error in value estimation. |
Large Language Models (LLMs) have achieved excellent performances in various tasks. However, fine-tuning an LLM requires extensive supervision. Human, on the other hand, may improve their reasoning abilities by self-thinking without external inputs. In this work, we demonstrate that an LLM is also capable of self-improving with only unlabeled datasets. We use a pre-trained LLM to generate “high-confidence” rationale-augmented answers for unlabeled questions using Chain-of-Thought prompting and self-consistency, and fine-tune the LLM using those self-generated solutions as target outputs. We show that our approach improves the general reasoning ability of a 540B-parameter LLM (74.4%→82.1% on GSM8K, 78.2%→83.0% on DROP, 90.0%→94.4% on OpenBookQA, and 63.4%→67.9% on ANLI-A3) and achieves state-of-the-art-level performance, without any ground truth label. We conduct ablation studies and show that finetuning on reasoning is critical for self-improvement. | Improving Reasoning Ability of Large Language Models in An Unsupervised Fashion |
We have recently seen great progress in learning interpretable music representations, ranging from basic factors, such as pitch and timbre, to high-level concepts, such as chord, texture and melody contour. However, most methods rely heavily on music domain knowledge and it remains an open question how to learn interpretable and disentangled representations using inductive biases that are more general. In this study, we use \textit{physical symmetry} as a self-consistency constraint on the latent space. Specifically, it requires the prior model that characterises the dynamics of the latent states to be \textit{equivariant} with respect to a certain group transformation (say, translation and rotation). We show that our model can learn \textit{linear} pitch factor (that agrees with human music perception) as well as pitch-timbre disentanglement from unlabelled monophonic music audio. In addition, the same methodology can be applied to computer vision, learning the 3D Cartesian space as well as space-colour disentanglement from a simple moving object shot by a single fix camera. Furthermore, applying physical symmetry to the prior model naturally leads to \textit{representation augmentation}, a new learning technique which helps improve sample efficiency. | This study uses physics symmetry as an effective inductive bias to learn interpretable representations from time-series data in a self-supervised fashion. |
This work introduces DiGress, a discrete denoising diffusion model for generating graphs with categorical node and edge attributes.
Our model utilizes a discrete diffusion process that progressively edits graphs with noise, through the process of adding or removing edges and changing the categories.
A graph transformer network is trained to revert this process, simplifying the problem of distribution learning over graphs into a sequence of node and edge classification tasks.
We further improve sample quality by introducing a Markovian noise model that preserves the marginal distribution of node and edge types during diffusion, and by incorporating auxiliary graph-theoretic features.
A procedure for conditioning the generation on graph-level features is also proposed.
DiGress achieves state-of-the-art performance on molecular and non-molecular datasets, with up to 3x validity improvement on a planar graph dataset.
It is also the first model to scale to the large GuacaMol dataset containing 1.3M drug-like molecules without the use of molecule-specific representations. | We propose a discrete denoising diffusion model for generating graphs with categorical node and edge attributes. It is state-of-the-art on both abstract and molecular datasets. |
Optimal transport aligns samples across distributions by minimizing the transportation cost between them, e.g., the geometric distances. Yet, it ignores coherence structure in the data such as clusters, does not handle outliers well, and cannot integrate new data points. To address these drawbacks, we propose InfoOT, an information-theoretic extension of optimal transport that maximizes the mutual information between domains while minimizing geometric distances. The resulting objective can still be formulated as a (generalized) optimal transport problem, and can be efficiently solved by projected gradient descent. This formulation yields a new projection method that is robust to outliers and generalizes to unseen samples. Empirically, InfoOT improves the quality of alignments across benchmarks in domain adaptation, cross-domain retrieval, and single-cell alignment. | We present InfoOT, an information-theoretic extension of optimal transport that combines the geometry with the mutual information between domains. |
The performance of acoustic models degrades notably in noisy environments. Speech enhancement (SE) can be used as a front-end strategy to aid automatic speech recognition (ASR) systems. However, existing training objectives of SE methods are not fully effective at integrating speech-text and noise-clean paired data for training toward unseen ASR systems. In this study, we propose a general denoising framework, D4AM, for various downstream acoustic models. Our framework fine-tunes the SE model with the backward gradient according to a specific acoustic model and the corresponding classification objective. In addition, our method aims to consider the regression objective as an auxiliary loss to make the SE model generalize to other unseen acoustic models. To jointly train an SE unit with regression and classification objectives, D4AM uses an adjustment scheme to directly estimate suitable weighting coefficients rather than undergoing a grid search process with additional training costs. The adjustment scheme consists of two parts: gradient calibration and regression objective weighting. The experimental results show that D4AM can consistently and effectively provide improvements to various unseen acoustic models and outperforms other combination setups. Specifically, when evaluated on the Google ASR API with real noisy data completely unseen during SE training, D4AM achieves a relative WER reduction of 24.65% compared with the direct feeding of noisy input. To our knowledge, this is the first work that deploys an effective combination scheme of regression (denoising) and classification (ASR) objectives to derive a general pre-processor applicable to various unseen ASR systems. Our code is available at https://github.com/ChangLee0903/D4AM. | We propose a general denoising framework for various downstream acoustic models (D4AM) by adopting an effective joint training scheme with the regression (denoising) objective and the classification (ASR) objective. |
Learning policies from previously recorded data is a promising direction for real-world robotics tasks, as online learning is often infeasible. Dexterous manipulation in particular remains an open problem in its general form. The combination of offline reinforcement learning with large diverse datasets, however, has the potential to lead to a breakthrough in this challenging domain analogously to the rapid progress made in supervised learning in recent years. To coordinate the efforts of the research community toward tackling this problem, we propose a benchmark including: i) a large collection of data for offline learning from a dexterous manipulation platform on two tasks, obtained with capable RL agents trained in simulation; ii) the option to execute learned policies on a real-world robotic system and a simulation for efficient debugging. We evaluate prominent open-sourced offline reinforcement learning algorithms on the datasets and provide a reproducible experimental setup for offline reinforcement learning on real systems. | We propose new robotics datasets for dexterous manipulation and benchmark offline RL algorithms on them. |
Numerically solving partial differential equations (PDEs) often entails spatial and temporal discretizations. Traditional methods (e.g., finite
difference, finite element, smoothed-particle hydrodynamics) frequently adopt explicit spatial discretizations, such as grids, meshes, and point clouds, where each degree-of-freedom corresponds to a location in space. While these explicit spatial correspondences are intuitive to model and understand, these representations are not necessarily optimal for accuracy, memory-usage, or adaptivity. In this work, we explore implicit neural representation as an alternative spatial discretization, where spatial information is implicitly stored in the neural network weights. With implicit neural spatial representation, PDE-constrained time-stepping translates into updating neural network weights, which naturally integrates with commonly adopted optimization time integrators. We validate our approach on a variety of classic PDEs with examples involving large elastic deformations, turbulent fluids, and multiscale phenomena. While slower to compute than traditional representations, our approach exhibits higher accuracy, lower memory consumption, and dynamically adaptive allocation of degrees of freedom without complex remeshing. | We replace traditional PDE solvers' spatial representations (e.g., grid, mesh, and point cloud) with a neural spatial representation. |
We consider a class of structured learning problems on arborescence (i.e., the directed spanning tree) from the input graph. The key step involved in this problem is predicting the minimal weight arborescence (MWA) from the learned model. In literature, there are two lines of research for predicting MWA: the Chu-Liu Edmonds (CLE) and the Lovasz methods. The CLE method is easy to implement while it takes $\mathcal{O}(n)$ cycle contractions. Here $n$ is the graph size. The Lovasz method reduces to the multi-pair shortest path (MPSP) problem and takes only $\mathcal{O}(\log n)$ contractions. Nevertheless, in the CPU setting, MPSP has the same time complexity as finding MWA. The Lovasz method only attains time efficiency under a sufficient GPU setting. Both the aforementioned methods are painfully slow for large-scale learning tasks. In this research, we find the general MPSP problem can be simplified when working with machine learning models. This is because the learning model predicts edge weights for all pairs of vertices and the graph we process is always complete. Therefore, we only need to handle those paths that directly enter every weakly connected component (WCC) while the classic Lovasz method needs to handle all possible paths. This allows us to propose LAzy LoVAz (Lava) method that enjoys $\mathcal{O}(\log n)$ contractions as well as efficient performance in both CPU and GPU settings. In experiments, we consider synthetic datasets and two real-world learning tasks, i.e., graph-based dependency parsing and unsupervised parsing on ListOps. The empirical results exhibit important gains of our Lava method to the classic CLE and Lovasz methods, that Lava boosts the training time for arborescence learning tasks. | We propose an efficient algorithm for predicting minimum weight arborescence, that can boost the training and inference time for arborescence learning tasks. |
Bayesian neural networks (BNNs) are touted for robustness under data drift, resilience to overfitting and catastrophic forgetting whilst also producing actionable uncertainty estimates. In variational inference, these elegant properties are contingent on the expressivity of the variational approximation. Posteriors over parameters of large models are usually multimodal and highly correlated and hence cannot be well-approximated by simple, prescribed densities. We posit implicit variational distributions specified using differentiable generators are more flexible and propose a novel bound for training BNNs using such approximations (amortized neural samplers). The proposed bound uses an approximation of the variational distribution's entropy by locally linearising the generator. Unlike existing works, our method does not require a discriminator network and moves away from an unfavourable adversarial objective. Our formulation resembles normalizing flows but does not necessitate invertibility of the generator. Moreover, we use a differentiable numerical lower bound on the Jacobians of the generator, mitigating computational concerns. We report log-likelihoods on UCI datasets competitive with deep ensembles and test our method on out-of-distribution benchmarks. | A novel bound for training implicit variational approximations for Bayesian Neural Networks |
Positional encodings are used to inject word-order information into transformer-based language models. While they can significantly enhance the quality of sentence representations, their specific contribution to language models are not fully understood, especially given recent findings that building natural-language understanding from language models with positional encodings is insensitive to word order. In this work, we investigate the role of positional encodings systematically. (1) We uncover the core function of existing positional encodings is to symmetrically combine local units by identifying two common properties, Locality, and Symmetry. (2) We reveal that positional and contextual encodings play a distinct role in understanding sentences. (3) Based on these findings, we propose a simplified new method to inject positional information into such models. Empirical studies demonstrate that this method can improve the performance of the BERT-based model on 10 downstream tasks. We hope these new probing results and findings can shed light on how to design and inject positional encodings into language models.
| In this work, we investigate the role of positional encodings systematically. |
Embedding discrete solvers as differentiable layers has given modern deep learning architectures combinatorial expressivity and discrete reasoning capabilities. The derivative of these solvers is zero or undefined, therefore a meaningful replacement is crucial for effective gradient-based learning. Prior works rely on smoothing the solver with input perturbations, relaxing the solver to continuous problems, or interpolating the loss landscape with techniques that typically require additional solver calls, introduce extra hyper-parameters, or compromise performance. We propose a principled approach to exploit the geometry of the discrete solution space to treat the solver as a negative identity on the backward pass and further provide a theoretical justification. Our experiments demonstrate that such a straightforward hyper-parameter-free approach is able to compete with previous more complex methods on numerous experiments such as backpropagation through discrete samplers, deep graph matching, and image retrieval. Furthermore, we substitute the previously proposed problem-specific and label-dependent margin with a generic regularization procedure that prevents cost collapse and increases robustness. | We propose a simple alternative for differentiating through combinatorial solvers with linear objectives, that is on par with SoTA, has no hyperparameters, and is more robust to perturbations. |
Pretrained Language Models (PLMs) such as BERT and its variants have achieved remarkable success in natural language processing. To date, the interpretability of PLMs has primarily relied on the attention weights in their self-attention layers. However, these attention weights only provide word-level interpretations, failing to capture higher-level structures, and are therefore lacking in readability and intuitiveness. In this paper, we propose a hierarchical Bayesian deep learning model, dubbed continuous latent Dirichlet allocation (CLDA), to go beyond word-level interpretations and provide concept-level interpretations. Our CLDA is compatible with any attention-based PLMs and can work as either (1) an interpreter which interprets model predictions at the concept level without any performance sacrifice or (2) a regulator which is jointly trained with PLMs during finetuning to further improve performance. Experimental results on various benchmark datasets show that our approach can successfully provide conceptual interpretation and performance improvement for state-of-the-art PLMs. | We propose a hierarchical Bayesian deep learning model to provide concept-level interpretations of pretrained language models. |
A problem with using the Gaussian distribution as a prior for the variational autoencoder (VAE) is that the set on which Gaussians have high probability density is small as the latent dimension increases. This is an issue because VAEs try to attain both a high likelihood with respect to a prior distribution and at the same time, separation between points for better reconstruction. Therefore, a small volume in the high-density region of the prior is problematic because it restricts the separation of latent points. To ameliorate this, we propose a simple generalization of the Gaussian distribution, called the tilted Gaussian, which has a maximum probability density occurring on a sphere instead of a single point. The tilted Gaussian has exponentially more volume in high-density regions than the standard Gaussian as a function of the distribution dimension. We empirically demonstrate that this simple change in the prior distribution improves VAE performance on the task of detecting unsupervised out-of-distribution (OOD) samples. We also introduce a new OOD testing procedure, called the Will-It-Move test, where the tilted Gaussian achieves remarkable OOD performance. | A generalization of the Gaussian distribution improves the performance of out-of-distribution detection with variational autoencoders. |
The success of deep learning heavily relies on large-scale data with comprehensive labels, which is more expensive and time-consuming to fetch in 3D compared to 2D images or natural languages. This promotes the potential of utilizing models pretrained with data more than 3D as teachers for cross-modal knowledge transferring. In this paper, we revisit masked modeling in a unified fashion of knowledge distillation, and we show that foundational Transformers pretrained with 2D images or natural languages can help self-supervised 3D representation learning through training Autoencoders as Cross-Modal Teachers (ACT). The pretrained Transformers are transferred as cross-modal 3D teachers using discrete variational autoencoding self-supervision, during which the Transformers are frozen with prompt tuning for better knowledge inheritance. The latent features encoded by the 3D teachers are used as the target of masked point modeling, wherein the dark knowledge is distilled to the 3D Transformer students as foundational geometry understanding. Our ACT pretrained 3D learner achieves state-of-the-art generalization capacity across various downstream benchmarks, e.g., 88.21% overall accuracy on ScanObjectNN. Codes have been released at https://github.com/RunpeiDong/ACT. | This paper shows that pretrained 2D image Transformers can help self-supervised 3D representation learning by training autoencoders as cross-modal teachers. |
We introduce Joint Multidimensional Scaling, a novel approach for unsupervised manifold alignment, which maps datasets from two different domains, without any known correspondences between data instances across the datasets, to a common low-dimensional Euclidean space. Our approach integrates Multidimensional Scaling (MDS) and Wasserstein Procrustes analysis into a joint optimization problem to simultaneously generate isometric embeddings of data and learn correspondences between instances from two different datasets, while only requiring intra-dataset pairwise dissimilarities as input. This unique characteristic makes our approach applicable to datasets without access to the input features, such as solving the inexact graph matching problem. We propose an alternating optimization scheme to solve the problem that can fully benefit from the optimization techniques for MDS and Wasserstein Procrustes. We demonstrate the effectiveness of our approach in several applications, including joint visualization of two datasets, unsupervised heterogeneous domain adaptation, graph matching, and protein structure alignment. The implementation of our work is available at https://github.com/BorgwardtLab/JointMDS. | A novel approach for unsupervised manifold alignment that only requires intra-domain pairwise dissimilarities as input. |
Given a visual scene, humans have strong intuitions about how a scene can evolve over time under given actions. The intuition, often termed visual intuitive physics, is a critical ability that allows us to make effective plans to manipulate the scene to achieve desired outcomes without relying on extensive trial and error. In this paper, we present a framework capable of learning 3D-grounded visual intuitive physics models purely from unlabeled images. Our method is composed of a conditional Neural Radiance Field (NeRF)-style visual frontend and a 3D point-based dynamics prediction backend, in which we impose strong relational and structural inductive bias to capture the structure of the underlying environment. Unlike existing intuitive point-based dynamics works that rely on the supervision of dense point trajectory from simulators, we relax the requirements and only assume access to multi-view RGB images and (imperfect) instance masks. This enables the proposed model to handle scenarios where accurate point estimation and tracking are hard or impossible. We evaluate the models on three challenging scenarios involving fluid, granular materials, and rigid objects, where standard detection and tracking methods are not applicable. We show our model can make long-horizon future predictions by learning from raw images and significantly outperforms models that do not employ an explicit 3D representation space. We also show that, once trained, our model can achieve strong generalization in complex scenarios under extrapolate settings. | An intuitive physics model with explicit 3D and compositional structures learned from multi-view videos. The learned model can handle complicated objects (e.g., fluid, rigid objects, granular materials) and perform extrapolated generalization. |
Transformers have quickly shined in the computer vision world since the emergence of Vision Transformers (ViTs). The dominant role of convolutional neural networks (CNNs) seems to be challenged by increasingly effective transformer-based models. Very recently, a couple of advanced convolutional models strike back with large kernels motivated by the local-window attention mechanism, showing appealing performance and efficiency. While one of them, i.e. RepLKNet, impressively manages to scale the kernel size to 31x31 with improved performance, the performance starts to saturate as the kernel size continues growing, compared to the scaling trend of advanced ViTs such as Swin Transformer. In this paper, we explore the possibility of training extreme convolutions larger than 31x31 and test whether the performance gap can be eliminated by strategically enlarging convolutions. This study ends up with a recipe for applying extremely large kernels from the perspective of sparsity, which can smoothly scale up kernels to 61x61 with better performance. Built on this recipe, we propose Sparse Large Kernel Network (SLaK), a pure CNN architecture equipped with sparse factorized 51x51 kernels that can perform on par with or better than state-of-the-art hierarchical Transformers and modern ConvNet architectures like ConvNeXt and RepLKNet, on ImageNet classification as well as a wide range of downstream tasks including semantic segmentation on ADE20K, object detection on PASCAL VOC 2007, and object detection/segmentation on MS COCO. Codes are available at https://github.com/VITA-Group/SLaK. | We propose Sparse Large Kernel Network (SLaK), a pure CNN architectures equipped with 51x51 kernels that can perform on par with or better than the state-of-the-art hierarchical Transformers and modern ConvNets. |
In real-world multi-agent reinforcement learning (MARL) applications, agents may not have perfect state information (e.g., due to inaccurate measurement or malicious attacks), which challenges the robustness of agents' policies. Though robustness is getting important in MARL deployment, little prior work has studied state uncertainties in MARL, neither in problem formulation nor algorithm design. Motivated by this robustness issue, we study the problem of MARL with state uncertainty in this work. We provide the first attempt to the theoretical and empirical analysis of this challenging problem. We first model the problem as a Markov Game with state perturbation adversaries (MG-SPA), and introduce Robust Equilibrium as the solution concept. We conduct fundamental analysis regarding MG-SPA and give conditions under which such an equilibrium exists. Then we propose a robust multi-agent Q-learning (RMAQ) algorithm to find such an equilibrium, with convergence guarantees. To handle high-dimensional state-action space, we design a robust multi-agent actor-critic (RMAAC) algorithm based on an analytical expression of the policy gradient derived in the paper. Our experiments show that the proposed RMAQ algorithm converges to the optimal value function; our RMAAC algorithm outperforms several MARL methods that do not consider the state uncertainty in several multi-agent environments. | fundamental research about robust multi-agent reinforcement learning with state uncertainty |
Adversarial attack serves as a major challenge for neural network models in NLP, which precludes the model's deployment in safety-critical applications. A recent line of work, detection-based defense, aims to distinguish adversarial sentences from benign ones. However, {the core limitation of previous detection methods is being incapable of giving correct predictions on adversarial sentences unlike defense methods from other paradigms.} To solve this issue, this paper proposes TextShield: (1) we discover a link between text attack and saliency information, and then we propose a saliency-based detector, which can effectively detect whether an input sentence is adversarial or not. (2) We design a saliency-based corrector, which converts the detected adversary sentences to benign ones. By combining the saliency-based detector and corrector, TextShield extends the detection-only paradigm to a detection-correction paradigm, thus filling the gap in the existing detection-based defense. Comprehensive experiments show that (a) TextShield consistently achieves higher or comparable performance than state-of-the-art defense methods across various attacks on different benchmarks. (b) our saliency-based detector outperforms existing detectors for detecting adversarial sentences. | A defense that extends adversarial detection paradigm in NLP |
Differentially private deep learning has recently witnessed advances in computational efficiency and privacy-utility trade-off. We explore whether further improvements along the two axes are possible and provide affirmative answers leveraging two instantiations of \emph{group-wise clipping}. To reduce the compute time overhead of private learning, we show that \emph{per-layer clipping}, where the gradient of each neural network layer is clipped separately, allows clipping to be performed in conjunction with backpropagation in differentially private optimization. This results in private learning that is as memory-efficient and almost as fast per training update as non-private learning for many workflows of interest. While per-layer clipping with constant thresholds tends to underperform standard flat clipping, per-layer clipping with adaptive thresholds matches or outperforms flat clipping under given training epoch constraints, hence attaining similar or better task performance within less wall time. To explore the limits of scaling (pretrained) models in differentially private deep learning, we privately fine-tune the 175 billion-parameter GPT-3. We bypass scaling challenges associated with clipping gradients that are distributed across multiple devices with \emph{per-device clipping} that clips the gradient of each model piece separately on its host device. Privately fine-tuning GPT-3 with per-device clipping achieves a task performance at $\epsilon=1$ better than what is attainable by non-privately fine-tuning the largest GPT-2 on a summarization task. | Explore the limit of the efficiency of DP-SGD with group-wise clipping |
What is an image, and how to extract latent features?
Convolutional Networks (ConvNets) consider an image as organized pixels in a rectangular shape and extract features via convolutional operation in a local region; Vision Transformers (ViTs) treat an image as a sequence of patches and extract features via attention mechanism in a global range. In this work, we introduce a straightforward and promising paradigm for visual representation, which is called Context Clusters. Context clusters (CoCs) view an image as a set of unorganized points and extract features via a simplified clustering algorithm. In detail, each point includes the raw feature (e.g., color) and positional information (e.g., coordinates), and a simplified clustering algorithm is employed to group and extract deep features hierarchically. Our CoCs are convolution- and attention-free, only relying on clustering algorithm for spatial interaction. Owing to the simple design, we show CoCs endow gratifying interpretability via the visualization of the clustering process.
Our CoCs aim at providing a new perspective on image and visual representation, which may enjoy broad applications in different domains and exhibit profound insights. Even though we are not targeting SOTA performance, COCs still achieve comparable or even better performance than ConvNets or ViTs on several benchmarks. | We introduce Context Cluster, a new paradigm that considers an image as a set of point and employs clustering method for feature extraction. |
Federated learning (FL) emerged as a novel machine learning setting that enables collaboratively training deep models on decentralized private data. Due to the heterogeneity (non-iidness) of the decentralized data, FL methods (e.g. FedAvg) suffers from unstable and slow convergence. Recent works explain the non-iid problem in FL as the client drift, and deal with it by enforcing regularization at local updates. However, these works neglect the heterogeneity among different communication rounds: the data of sampled candidates at different communication rounds are also of non-iid distribution, and we term it as period drift, which as well as client drift can lead to aggregation bias that degrade convergence. To deal with it, we propose a novel aggregation strategy, named FedPA, that uses a Parameterized Aggregator, as an alternative of averaging. We frame FedPA within a meta-learning setting, and formulates the aggregator as a meta-learner, to learn to aggregate the model parameters of clients. FedPA can directly learn the aggregation bias and well calibrate and control the direction of aggregated parameters to a better direction towards the optimum. Experiments show that FedPA can achieve competitive performances compared with conventional baselines. | Our idea is to learn an aggregator to debias aggregation to calibrate and control the direction of aggregated parameters to deal with both client drift and period drift. |
Much of the data we encounter in the real world can be represented as directed graphs. In this work, we introduce a general family of representations for directed graphs through connected time-oriented Lorentz manifolds, called "spacetimes" in general relativity. Spacetimes intrinsically contain a causal structure that indicates whether or not there exists a causal or even chronological order between points of the manifold, called events. This chronological order allows us to naturally represent directed edges via imposing the correct ordering when the nodes are embedded as events in the spacetime. Previous work in machine learning only considers embeddings lying on the simplest Lorentz manifold or does not exploit the connection between Lorentzian pre-length spaces and directed graphs. We introduce a well-defined approach to map data onto a general family of spacetimes. We empirically evaluate our framework in the tasks of hierarchy extraction of undirected graphs, directed link prediction and representation of directed graphs. | Representation of directed graphs by exploiting the causal structure of spacetimes via Lorentzian pre-length spaces |
In this paper we motivate the causal mechanisms behind sample selection induced collider bias (selection collider bias) that can cause Large Language Mod- els (LLMs) to learn unconditional dependence between entities that are unconditionally independent in the real world. We show that selection collider bias can become amplified in underspecified learning tasks, and although difficult to overcome, we describe a method to exploit the resulting spurious correlations for determination of when a model may be uncertain about its prediction. We demonstrate an uncertainty metric that matches human uncertainty in tasks with gender pronoun underspecification on an extended version of the Winogender Schemas evaluation set, and we provide online demos where users can evaluate spurious correlations and apply our uncertainty metric to their own texts and models. Finally, we generalize our approach to address a wider range of prediction tasks. | Using causal inference methods, we explain and demonstrate how sample selection bias causes spurious correlations during training, and how those spurious correlations can be used to classify prediction tasks as underspecified during inference. |
A neural ordinary differential equation (ODE) is a relation between an unknown function and its derivatives, where the ODE is parameterized by a neural network. Therefore, to obtain a solution to a neural ODE requires a solver that performs numerical integration. Dopri5 is one of the most popular neural ODE solvers and also the default solver in torchdiffeq, a PyTorch library of ODE solvers. It is an adaptive step size solver based on the Runge-Kutta (RK) numerical methods. These methods rely on estimation of the local truncation error to select and adjust integration step size, which determines the numerical stability of the solution. A step size that is too large leads to numerical instability, while a step size that is too small may cause the solver to take unnecessarily many steps, which is computationally expensive and may even cause rounding error build up. Therefore, accurate local truncation error estimation is paramount for choosing an appropriate step size to obtain an accurate, numerically stable, and fast solution to the ODE. In this paper we propose a novel local truncation error approximation that is the first to consider solutions of four different RK orders to obtain a more reliable error estimate. This leads to a novel solver S-SOLVER (Stable Solver), which is more numerically stable; and therefore accurate. We demonstrate S-SOLVER's competitive performance in experiments on image recognition with ODE-Net, learning hamiltonian dynamics with Symplectic ODE-Net, and continuous normalizing flows (CNF). | We propose a neural ODE adaptive step size solver that is more numerically stable thanks to novel, more reliable local truncation error estimation. |
When trained on language data, do transformers learn some arbitrary computation that utilizes the full capacity of the architecture or do they learn a simpler, tree-like computation, hypothesized to underlie compositional meaning systems like human languages? There is an apparent tension between compositional accounts of human language understanding, which are based on a restricted bottom-up computational process, and the enormous success of neural models like transformers, which can route information arbitrarily between different parts of their input. One possibility is that these models, while extremely flexible in principle, in practice learn to interpret language hierarchically, ultimately building sentence representations close to those predictable by a bottom-up, tree-structured model. To evaluate this possibility, we describe an unsupervised and parameter-free method to \emph{functionally project} the behavior of any transformer into the space of tree-structured networks. Given an input sentence, we produce a binary tree that approximates the transformer's representation-building process and a score that captures how ``tree-like'' the transformer's behavior is on the input. While calculation of this score does not require training any additional models, it provably upper-bounds the fit between a transformer and any tree-structured approximation. Using this method, we show that transformers for three different tasks become more tree-like over the course of training, in some cases unsupervisedly recovering the same trees as supervised parsers. These trees, in turn, are predictive of model behavior, with more tree-like models generalizing better on tests of compositional generalization. | We provide a method to functionally project a transformer into the space of tree structured models and use it to uncover intrinsic compositionality of transformers trained on language data. |
Many recent works have shown trainability plays a central role in neural network pruning -- unattended broken trainability can lead to severe under-performance and unintentionally amplify the effect of retraining learning rate, resulting in biased (or even misinterpreted) benchmark results. This paper introduces trainability preserving pruning (TPP), a scalable method to preserve network trainability against pruning, aiming for improved pruning performance and being more robust to retraining hyper-parameters (e.g., learning rate). Specifically, we propose to penalize the gram matrix of convolutional filters to decorrelate the pruned filters from the retained filters. In addition to the convolutional layers, per the spirit of preserving the trainability of the whole network, we also propose to regularize the batch normalization parameters (scale and bias). Empirical studies on linear MLP networks show that TPP can perform on par with the oracle trainability recovery scheme. On nonlinear ConvNets (ResNet56/VGG19) on CIFAR10/100, TPP outperforms the other counterpart approaches by an obvious margin. Moreover, results on ImageNet-1K with ResNets suggest that TPP consistently performs more favorably against other top-performing structured pruning approaches. Code: https://github.com/MingSun-Tse/TPP. | We present a new filter pruning approach that effectively preserves trainability during pruning with encouraging performance. |
While deep learning has achieved great success in various fields, a large amount of memory is necessary to train deep neural networks, which hinders the development of massive state-of-the-art models. The reason is the conventional learning rule, backpropagation, should temporarily store input activations of all the layers in the network. To overcome this, recent studies suggested various memory-efficient implementations of backpropagation. However, those approaches incur computational overhead due to the recomputation of activations, slowing down neural network training. In this work, we propose a new learning rule which significantly reduces memory requirements while closely matching the performance of backpropagation. The algorithm combines auxiliary activation with output activation during forward propagation, while only auxiliary activation is used during backward propagation instead of actual input activation to reduce the amount of data to be temporarily stored. We mathematically show that our learning rule can reliably train the networks whose loss landscape is convex if the auxiliary activation satisfies certain conditions. Based on this observation, we suggest candidates of auxiliary activation that satisfy those conditions. Experimental results confirm that the proposed learning rule achieves competitive performance compared to backpropagation in various models such as ResNet, Transformer, BERT, ViT, and MLP-Mixer. | The proposed learning rule reduces training memory requirements without reduction in training speed while achieving high performance close to backpropagation. |
State-space models (SSMs) perform predictions by learning the underlying dynamics of observed sequence. We propose a new SSM in both high and low dimensional observation space, which utilizes Bayesian filtering-smoothing to model system’s dynamics more accurately than RNN-based SSMs and can be learned in an end-to-end manner. The designed architecture, which we call the Gated Inference Network (GIN), is able to integrate the uncertainty estimates and learn the complicated dynamics of the system that enables us to perform estimation and imputation tasks in both data presence and absence. The proposed model uses the GRU cells into its structure to complete the data flow, while avoids expensive computations and potentially unstable matrix inversions. The GIN is able to deal with any time-series data and gives us a strong robustness to handle the observational noise. In the numerical experiments, we show that the GIN reduces the
uncertainty of estimates and outperforms its counterparts , LSTMs, GRUs and variational approaches. | A structure, using Bayesian properties and non-linearity in its design, is introduced that can learn complex state-spaces. |
Since the development of self-supervised visual representation learning from contrastive learning to masked image modeling (MIM), there is no significant difference in essence, that is, how to design proper pretext tasks for vision dictionary look-up. MIM recently dominates this line of research with state-of-the-art performance on vision Transformers (ViTs), where the core is to enhance the patch-level visual context capturing of the network via denoising auto-encoding mechanism. Rather than tailoring image tokenizers with extra training stages as in previous works, we unleash the great potential of contrastive learning on de- noising auto-encoding and introduce a pure MIM method, ConMIM, to produce simple intra-image inter-patch contrastive constraints as the sole learning objectives for masked patch prediction. We further strengthen the denoising mechanism with asymmetric designs, including image perturbations and model progress rates, to improve the network pre-training. ConMIM-pretrained models with various scales achieve competitive results on downstream image classification, semantic segmentation, object detection, and instance segmentation tasks, e.g., on ImageNet-1K classification, we achieve 83.9% top-1 accuracy with ViT-Small and 85.3% with ViT-Base without extra data for pre-training. Code will be available at https://github.com/TencentARC/ConMIM.
| We first treat masked patch prediction as denoising contrastive learning in self-supervised image pre-training, achieving state-of-the-art results. |
Recurrent neural network (RNN) and self-attention mechanism (SAM) are the de facto methods to extract spatial-temporal information for temporal graph learning. Interestingly, we found that although both RNN and SAM could lead to a good performance, in practice neither of them is always necessary. In this paper, we propose GraphMixer, a conceptually and technically simple architecture that consists of three components: (1) a link-encoder that is only based on multi-layer perceptrons (MLP) to summarize the information from temporal links, (2) a node-encoder that is only based on neighbor mean-pooling to summarize node information, and (3) an MLP-based link classifier that performs link prediction based on the outputs of the encoders. Despite its simplicity, GraphMixer attains an outstanding performance on temporal link prediction benchmarks with faster convergence and better generalization performance. These results motivate us to rethink the importance of simpler model architecture. | This paper propose a conceptually and technically simple method for temporal graph link prediction |
We propose a novel clustering mechanism based on an incompatibility property between subsets of data that emerges during model training. This mechanism partitions the dataset into subsets that generalize only to themselves, i.e., training on one subset does not improve performance on the other subsets. Leveraging the interaction between the dataset and the training process, our clustering mechanism partitions datasets into clusters that are defined by—and therefore meaningful to—the objective of the training process.
We apply our clustering mechanism to defend against data poisoning attacks, in which the attacker injects malicious poisoned data into the training dataset to affect the trained model's output. Our evaluation focuses on backdoor attacks against deep neural networks trained to perform image classification using the GTSRB and CIFAR-10 datasets. Our results show that (1) these attacks produce poisoned datasets in which the poisoned and clean data are incompatible and (2) our technique successfully identifies (and removes) the poisoned data. In an end-to-end evaluation, our defense reduces the attack success rate to below 1% on 134 out of 165 scenarios, with only a 2% drop in clean accuracy on CIFAR-10 and a negligible drop in clean accuracy on GTSRB. | We observe that training with poisoned data does not improve clean accuracy (and vice-versa), and develop a defense that exploits this property. |
It is expensive to collect training data for every possible domain that a vision model may encounter when deployed. We instead consider how simply $\textit{verbalizing}$ the training domain (e.g.``photos of birds'') as well as domains we want to extend to but do not have data for (e.g.``paintings of birds'') can improve robustness. Using a multimodal model with a joint image and language embedding space, our method $\textit{LADS}$ learns a transformation of the image embeddings from the source domain to each target domain, while preserving task relevant information. Without using any images from the target domain, we show that over the $\textit{extended}$ domain containing both source and target, $\textit{LADS}$ outperforms standard fine-tuning and ensemble approaches over a suite of 4 benchmarks targeting domain adaptation and dataset bias. | Transforming multimodal embeddings with language improves accuracy on an unseen domain. |
Training a robust cooperative agent requires diverse partner agents. However, obtaining those agents is difficult. Previous works aim to learn diverse behaviors by changing the state-action distribution of agents. But, without information about the task's goal, the diversified agents are not guided to find other important, albeit sub-optimal, solutions: the agents might learn only variations of the same solution. In this work, we propose to learn diverse behaviors via policy compatibility. Conceptually, policy compatibility measures whether policies of interest can coordinate effectively. We theoretically show that incompatible policies are not similar. Thus, policy compatibility—which has been used exclusively as a measure of robustness—can be used as a proxy for learning diverse behaviors. Then, we incorporate the proposed objective into a population-based training scheme to allow concurrent training of multiple agents. Additionally, we use state-action information to induce local variations of each policy. Empirically, the proposed method consistently discovers more solutions than baseline methods across various multi-goal cooperative environments. Finally, in multi-recipe Overcooked, we show that our method produces populations of behaviorally diverse agents, which enables generalist agents trained with such a population to be more robust.
See our project page at https://bit.ly/marl-lipo
| We show that incompatible poclies are not similar. LIPO generates diverse cooperative partners by learning a population of incompatible policies. |
The robustness of machine learning algorithms to distributions shift is primarily discussed in the context of supervised learning (SL). As such, there is a lack of insight on the robustness of the representations learned from unsupervised methods, such as self-supervised learning (SSL) and auto-encoder based algorithms (AE), to distribution shift. We posit that the input-driven objectives of unsupervised algorithms lead to representations that are more robust to distribution shift than the target-driven objective of SL. We verify this by extensively evaluating the performance of SSL and AE on both synthetic and realistic distribution shift datasets. Following observations that the linear layer used for classification itself can be susceptible to spurious correlations, we evaluate the representations using a linear
head trained on a small amount of out-of-distribution (OOD) data, to isolate the robustness of the learned representations from that of the linear head. We also develop “controllable” versions of existing realistic domain generalisation datasets with adjustable degrees of distribution shifts. This allows us to study the robustness of different learning algorithms under versatile yet realistic distribution shift
conditions. Our experiments show that representations learned from unsupervised learning algorithms generalise better than SL under a wide variety of extreme as well as realistic distribution shifts. | Representations learned from self-supervised learning and auto-encoder based algorithms are surprisingly robust to distribution shift. |
Reinforcement Learning (RL) algorithms can solve challenging control problems directly from image observations, but they often require millions of environment interactions to do so. Recently, model-based RL algorithms have greatly improved sample-efficiency by concurrently learning an internal model of the world, and supplementing real environment interactions with imagined rollouts for policy improvement. However, learning an effective model of the world from scratch is challenging, and in stark contrast to humans that rely heavily on world understanding and visual cues for learning new skills. In this work, we investigate whether internal models learned by modern model-based RL algorithms can be leveraged to solve new, distinctly different tasks faster. We propose Model-Based Cross-Task Transfer (XTRA), a framework for sample-efficient online RL with scalable pretraining and finetuning of learned world models. By offline multi-task pretraining and online cross-task finetuning, we achieve substantial improvements over a baseline trained from scratch; we improve mean performance of model-based algorithm EfficientZero by 23%, and by as much as 71% in some instances. Project page: https://nicklashansen.github.io/xtra | We investigate the feasibility of pretraining and cross-task transfer in model-based RL, and improve sample-efficiency substantially over baselines on the Atari100k benchmark. |
A wide variety of model explanation approaches have been proposed in recent years, all guided by very different rationales and heuristics. In this paper, we take a new route and cast interpretability as a statistical inference problem. We propose a general deep probabilistic model designed to produce interpretable predictions. The model’s parameters can be learned via maximum likelihood, and the method can be adapted to any predictor network architecture, and any type of prediction problem. Our method is a case of amortized interpretability models, where a neural network is used as a selector to allow for fast interpretation at inference time. Several popular interpretability methods are shown to be particular cases of regularised maximum likelihood for our general model. We propose new datasets with ground truth selection which allow for the evaluation of the features importance map. Using these datasets, we show experimentally that using multiple imputation provides more reasonable interpretation. | We propose to embed any classification or regression model in a framework that casts interpretability as a maximum likelihood problem. |
We aim to learn a set of temporal logic rules to explain the occurrence of temporal events. Leveraging the temporal point process modeling and learning framework, the rule content and rule weights are jointly learned by maximizing the likelihood of the observed noisy event sequences. The proposed algorithm alternates between a master problem, where the rule weights are updated, and a subproblem, where a new rule is searched and included. The formulated master problem is convex and relatively easy to solve, whereas the subproblem requires searching the huge combinatorial rule predicate and relationship space. To tackle this challenge, we propose a neural search policy to learn to generate the new rule content as a sequence of actions. The policy parameters will be trained end-to-end using the reinforcement learning framework, where the reward signals can be efficiently queried by evaluating the subproblem objective. The trained policy can be used to generate new rules, and moreover, the well-trained policies can be directly transferred to other tasks to speed up the rule searching procedure in the new task. We evaluate our methods on both synthetic and real-world datasets, obtaining promising results. | We aim to learn a set of temporal logic rules to explain the temporal point processes. |
We currently do not have an understanding of semi-supervised learning (SSL) objectives such as pseudo-labelling and entropy minimization as log-likelihoods, which precludes the development of e.g. Bayesian SSL. Here, we note that benchmark image datasets such as CIFAR-10 are carefully curated, and we formulate SSL objectives as a log-likelihood in a generative model of data curation. We show that SSL objectives, from entropy minimization and pseudo-labelling, to state-of-the-art techniques similar to FixMatch can be understood as lower-bounds on our principled log-likelihood. We are thus able to introduce a Bayesian extension of SSL, which gives considerable improvements over standard SSL in the setting of 40 labelled points on CIFAR-10, with performance of $92.2\pm 0.3\%$ vs $88.6\%$ in the original FixMatch paper. Finally, our theory suggests that SSL is effective in part due to the statistical patterns induced by data curation. This provides an explanation of past results which show SSL performs better on clean datasets without any ``out of distribution'' examples. Confirming these results we find that SSL gave much larger performance improvements on curated than on uncurated data, using matched curated and uncurated datasets based on Galaxy Zoo 2. | We develop Bayesian semi-supervised learning, by showing that standard SSL objectives can be understood as lower bounds on a principled log-likelihood |
Edge devices provide inference on predictive tasks to many end-users. However, deploying deep neural networks that achieve state-of-the-art accuracy on these devices is infeasible due to edge resource constraints. Nevertheless, cloud-only processing, the de-facto standard, is also problematic, since uploading large amounts of data imposes severe communication bottlenecks. We propose a novel end-to-end hybrid learning framework that allows the edge to selectively query only those hard examples that the cloud can classify correctly. Our framework optimizes over neural architectures and trains edge predictors and routing models so that the overall accuracy remains high while minimizing the overall latency. Training a hybrid learner is difficult since we lack annotations of hard edge-examples. We introduce a novel proxy supervision in this context and show that our method adapts seamlessly and near optimally across different latency regimes. On the ImageNet dataset, our proposed method deployed on a micro-controller unit exhibits $25\%$ reduction in latency compared to cloud-only processing while suffering no excess loss. | Low-complexity model(edge) performs poorly on large-scale tasks; For efficient inference, it must learn to identify examples that benefit by querying; it has to identify both hard-to-classify examples and those that the cloud model would misclassify. |
Numerical simulation of non-linear partial differential equations plays a crucial role in modeling physical science and engineering phenomena, such as weather, climate, and aerodynamics. Recent Machine Learning (ML) models trained on low-resolution spatio-temporal signals have shown new promises in capturing important dynamics in high-resolution signals, under the condition that the models can effectively recover the missing details. However, this study shows that significant information is often lost in the low-resolution down-sampled features. To address such issues, we propose a new approach, namely Temporal Stencil Modeling (TSM), which combines the strengths of advanced time-series sequence modeling (with the HiPPO features) and state-of-the-art neural PDE solvers (with learnable stencil modeling). TSM aims to recover the lost information from the PDE trajectories and can be regarded as a temporal generalization of classic finite volume methods such as WENO. Our experimental results show that TSM achieves the new state-of-the-art simulation accuracy for 2-D incompressible Navier-Stokes turbulent flows: it significantly outperforms the previously reported best results by 19.9% in terms of the highly-correlated duration time, and reduces the inference latency into 80%. We also show a strong generalization ability of the proposed method to various out-of-distribution turbulent flow settings. | We propose a novel Temporal Stencil Modeling (TSM) method for solving time-dependent PDEs in conservation form. |
While large language models (LLMs) have demonstrated impressive capabilities across tasks in language understanding and interactive decision making, their abilities for reasoning (e.g. chain-of-thought prompting) and acting (e.g. action plan generation) have primarily been studied as separate topics. In this paper, we explore the use of LLMs to generate both reasoning traces and task-specific actions in an interleaved manner, allowing for greater synergy between the two: reasoning traces help the model induce, track, and update action plans as well as handle exceptions, while actions allow it to interface with external sources, such as knowledge bases or environments, to gather additional information. We apply our approach, named ReAct, to a diverse set of language and decision making tasks and demonstrate its effectiveness over state-of-the-art baselines, as well as improved human interpretability and trustworthiness over methods without reasoning or acting components. Concretely, on question answering (HotpotQA) and fact verification (Fever), ReAct overcomes issues of hallucination and error propagation prevalent in chain-of-thought reasoning by interacting with a simple Wikipedia API, and generates human-like task-solving trajectories that are more interpretable than baselines without reasoning traces. On two interactive decision making benchmarks (ALFWorld and WebShop), ReAct outperforms imitation and reinforcement learning methods by an absolute success rate of 34% and 10% respectively, while being prompted with only one or two in-context examples. | We synergize reasoning and action taking in language models and make them more capable, versatile and interpretable. |
Demonstrations and natural language instructions are two common ways to specify and teach robots novel tasks. However, for many complex tasks, a demonstration or language instruction alone contains ambiguities, preventing tasks from being specified clearly. In such cases, a combination of both a demonstration and an instruction more concisely and effectively conveys the task to the robot than either modality alone. To instantiate this problem setting, we train a single multi-task policy on a few hundred challenging robotic pick-and-place tasks and propose DeL-TaCo (Joint Demo-Language Task Conditioning), a method for conditioning a robotic policy on task embeddings comprised of two components: a visual demonstration and a language instruction. By allowing these two modalities to mutually disambiguate and clarify each other during novel task specification, DeL-TaCo (1) substantially decreases the teacher effort needed to specify a new task and (2) achieves better generalization performance on novel objects and instructions over previous task-conditioning methods. To our knowledge, this is the first work to show that simultaneously conditioning a multi-task robotic manipulation policy on both demonstration and language embeddings improves sample efficiency and generalization over conditioning on either modality alone. | Conditioning robotic manipulation policies on both demonstrations and language instructions improves sample efficiency and generalization to novel tasks. |
Self-supervised learning (SSL) has become the predominant approach to training on large amounts of unlabeled data. New real-world APIs offer services to generate high-dimensional representations for given inputs based on SSL encoders with transformer architectures. Recent efforts highlight that it is possible to steal high-quality SSL encoders trained on convolutional neural networks. In this work, we are the first to extend this line of work to stealing and defending transformer-based encoders in both language and vision domains. We show that it is possible to steal transformer-based sentence embedding models solely using their returned representations and with 40x fewer queries than the number of victim's training data points. We also decrease the number of required stealing queries for the vision encoders by leveraging semi-supervised learning. Finally, to defend vision transformers against stealing attacks, we propose a defense technique that combines watermarking with dataset inference. Our method creates a unique encoder signature based on a private data subset that acts as a secret seed during training. By applying dataset inference on the seed, we can then successfully identify stolen transformers. | We perform attacks against transformer-based encoders and propose a new defense against extraction of vision transformers that combines watermarking with dataset inference. |
A disentangled representation encodes generative factors of data in a separable and compact pattern. Thus it is widely believed that such a representation format benefits downstream tasks. In this paper, we challenge the necessity of disentangled representation in downstream applications. Specifically, we show that dimension-wise disentangled representations are not necessary for downstream tasks using neural networks that take learned representations as input. We provide extensive empirical evidence against the necessity of disentanglement, covering multiple datasets, representation learning methods, and downstream network architectures. Moreover, our study reveals that the informativeness of representations best accounts for downstream performance. The positive correlation between informativeness and disentanglement explains the claimed usefulness of disentangled representations in previous works. | We show that dimension-wise disentangled representations are not necessary for downstream tasks using deep neural networks with learned representations as input. |
There are sequence generation tasks where the best order to generate the target sequence is not left-to-right. For example, an answer to the Sudoku game, a structured code like s-expression, and even a logical natural language answer where the analysis may be generated after the decision. We define the target sequences of those tasks as complex sequences. Obviously, a complex sequence should be constructed with multiple logical steps, and has dependencies among each part of itself (e.g. decisions depend on analyses). It's a great challenge for the classic left-to-right autoregressive generation system to generate complex sequences. Current approaches improve one-pass left-to-right generation on NLG tasks by generating different heuristic intermediate sequences in multiple stages. However, for complex sequences, the heuristic rules to break down them may hurt performance, and increase additional exposure bias. To tackle these challenges, we propose a PLM-friendly autoregressive self-boost refinement framework, CASR. When training, CASR inputs the predictions generated by the model itself at the previous refinement step (instead of those produced by heuristic rules). To find an optimal design, we also discuss model architecture, parameter efficiency and initialization strategy. By evaluating CASR on Sudoku, WebQSP, MTOP and KVRET through controlled experiments and empirical studies, we find that CASR produces high-quality outputs. CASR also improves Accuracy on Sudoku (70.93% --> 97.28%) and achieves state-of-the-art performance on KVRET with Micro F1 score (67.88% --> 70.00%). | CASR improves left-to-right autoregressive generation without heuristic intermediate sequences for complex answers via self-boost refinement |
Subgraph counting is the problem of determining the number of a given query graph in a large targe graph. Despite being a #P problem, subgraph counting is a crucial graph analysis method in domains ranging from biology and social science to risk management and software analysis. However, existing exact counting methods take combinatorially long runtime as target and query sizes increase. Existing approximate heuristic methods and neural approaches fall short in accuracy due to high label dynamic range, limited model expressive power, and inability to predict the distribution of subgraph counts in the target graph. Here we propose DeSCo, a neural deep subgraph counting framework, which aims to accurately predict the count and distribution of query graphs on any given target graph. DeSCo uses canonical partition to divide the large target graph into small neighborhood graphs and predict the canonical count objective on each neighborhood. The proposed partition method avoids missing or double-counting any patterns of the target graph. A novel subgraph-based heterogeneous graph neural network is then used to improve the expressive power. Finally, gossip correction improves counting accuracy via prediction propagation with learnable weights. Compared with state-of-the-art approximate heuristic and neural methods. DeSCo achieves 437x improvement in the mean squared error of count prediction and benefits from the polynomial runtime complexity.
| We propose DeSCo, a neural-based deep subgraph counting framework aims to accurately predict count of query graphs on any given target graph. |